{"id":"4a2f95d7-ec66-41e3-8b26-cb196e82d942","arxiv_id":"2504.17328","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The log-ratio distance on unit-area triangles is a complete Finsler metric of Thompson type; its geodesic, Finsler, and completeness structure extends to triangulated Euclidean surfaces, with an example showing forward incompleteness there.","lead":"This paper puts an asymmetric distance on the space of unit-area Euclidean triangles, shows it is a Finsler metric of Thompson type with explicit geodesics, extends the structure to surfaces with fixed triangulations, and develops a completeness theory for asymmetric metrics. It matters as a fully explicit Euclidean analogue of Thurston's stretch-metric theory, with a concrete example where the surface space fails forward completeness.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Corollary 4.8's bigeodesic construction leaves E(S,T): coordinate-wise geometric interpolation does not preserve the linear edge equalities, so the existence proof for geodesics and the Finsler structure on triangulated surfaces is invalid as written.","rationale":"The reader's weakest assumption was the unproved converse in Proposition 4.3, i.e. that the image of the face-coordinate parametrization is exactly the linear equalities per interior edge. That converse is actually true and can be proved by reconstructing each edge length as a sum of two face coordinates; it is not the most fragile point. The most concrete load-bearing gap I find is the proof of Corollary 4.8, which is not merely sketched but uses an invalid construction. Since E(S,T) is a convex cone, the true bigeodesic can be taken to be (a unit-area normalization of) the affine segment, so the statement is likely correct; however, the paper as written does not give this proof and the proof it does give is wrong. Proposition 4.10 and Theorem 4.12 rely on the existence of geodesics and are asserted without a valid proof, so the generalized Finsler claims are not fully established by the text. This reinforces the reader's CONDITIONAL verdict rather than overturning it: the mathematical results appear sound, but the written arguments in Sections 4-7 need correction and expansion. The Section 3 partial-derivative factor-of-2 error and the use of points outside T1 in the asymmetry example are additional correctness defects, but they do not threaten the central claims as much as the invalid geodesic construction does.","tokens_in":21267,"tokens_out":41147,"duration_ms":394080,"concrete_test":"Take a disc triangulated by two triangles sharing an interior edge 1, with the other edges labelled 2,3 and 4,5 respectively. Let A have edge lengths (5,12,13,12,13) and B have (5,11,14,13,12). The A-coordinate sums for edge 1 are equal for A and for B: A_2^1+A_3^1 = A_4^2+A_5^2 = 5 in both. Compute the coordinate-wise geometric interpolant at t=1/2: for face 1 the relevant sum is sqrt(3*4)+sqrt(2*1)=2√3+√2≈4.878, while for face 2 it is sqrt(3*2)+sqrt(2*3)=2√6≈4.899. The sums differ, so the interpolant leaves E(S,T). If unit-area scaling is applied first to A and B, the same discrepancy persists up to a common positive factor. This directly falsifies the construction in Corollary 4.8. As a control, check whether the affine path (1-t)A+tB, normalized to unit area, is a geodesic; if so, the theorem is true but the proof must be replaced.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Corollary 4.8 constructs a bigeodesic between two points A,B in E(S,T)_1 by setting α(t) = (A_{ijk}^{1-t} B_{ijk}^t) coordinate-wise. This path does not generally lie in E(S,T). E(S,T) is cut out inside the positive orthant by the linear equalities A_{jki}+A_{kij}=A_{j'k'i'}+A_{k'i'j'} for each interior edge i of the triangulation. Coordinate-wise geometric interpolation does not preserve sums: from u_0+v_0=u'_0+v'_0 and u_1+v_1=u'_1+v'_1 it does not follow that u_0^{1-t}u_1^t+v_0^{1-t}v_1^t = u'_0^{1-t}u'_1^t+v'_0^{1-t}v'_1^t. A concrete two-triangle example is given in the concrete_test below. Hence the path α(t) is not in E(S,T), so the subsequent normalization λ(t)α(t) cannot define a path in E(S,T)_1. This matters because Corollary 4.8 is the only supplied evidence for the existence of geodesics and bigeodesics in the triangulated-surface setting, and Proposition 4.10 / Theorem 4.12 (the Finsler realization of η(T)) are asserted by analogy with the single-triangle case without a valid geodesic-producing construction. The central claims for Sections 4-7 therefore rest on an argument that is, at minimum, incomplete. The theorem itself appears salvageable: the affine path (1-t)A+tB stays in the convex cone E(S,T) and the coordinate with maximal ratio B_i/A_i has the largest logarithmic derivative along it, so a bigeodesic exists; but that is not the proof given in the paper.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces an asymmetric distance η on the space T1 of unit-area marked Euclidean triangles, parametrized by semiperimeter-defect coordinates A_i = (a_j + a_k − a_i)/2, defined by η(X,Y) = log max_i(Y_i/X_i). It proves that η is a complete Finsler metric on T1, gives a box-Lipschitz equivalent formulation, characterizes geodesics by a coordinate-dominance condition, constructs bigeodesics, and gives the infinitesimal norm F(A,v) = max_i(v_i/A_i). It then extends the construction to E(S,T)_1, the space of unit-area singular Euclidean structures on a fixed triangulation, claiming the same geodesic, Finsler, and completeness properties, and develops a theory of completion for asymmetric metrics. A final section studies spaces of convex Euclidean polygons with several triangulations.","tokens_in":21444,"tokens_out":22857,"duration_ms":184691,"significance":"The single-triangle portion of the paper is clean and well executed; the metric is a non-symmetric Thompson-type distance in logarithm coordinates, and the explicit formulas for geodesics and the Finsler norm are useful and clearly motivated by Thurston's theory. If the triangulated-surface results are repaired as indicated below, the paper would provide a natural Euclidean analogue of aspects of Thurston's metric, together with a concrete forward-incomplete example (Example 7.7) that is both instructive and valuable. The completion theory for asymmetric metrics is a reasonable contribution, though the proof of one of its key steps needs repair.","major_comments":[{"comment":"The construction of the bigeodesic is invalid because the path α(t) = (A_{ijk}^{1−t} B_{ijk}^t) is defined coordinate-wise in the ambient space (R_+^*)^J, while E(S,T) is cut out by the linear equalities of Proposition 4.3. Coordinate-wise geometric interpolation does not preserve those linear equalities: from u_0+v_0 = u'_0+v'_0 and u_1+v_1 = u'_1+v'_1 it does not follow that u_0^{1−t}u_1^t + v_0^{1−t}v_1^t = u'_0^{1−t}u'_1^t + v'_0^{1−t}v'_1^t (for example, take u_0=v'_0=1, v_0=u'_0=2, u_1=v'_1=1, v_1=u'_1=8, where the sums match at t=0 and t=1 but differ at t=1/2). Hence λ(t)α(t) is not a path in E(S,T)_1, so Corollary 4.8 does not prove the existence of a bigeodesic, and the proofs of Proposition 4.10 and Theorem 4.13 that invoke it are unsupported. The statement itself is true and repairable: the affine path β(t) = (1−t)A + tB stays in E(S,T) because the defining equalities are linear, and if j maximizes B_j/A_j, then the logarithmic derivative of β_i is pointwise dominated by that of β_j, so the normalized affine path is a bigeodesic by Theorem 4.6. Please replace the construction and amend the proofs that depend on it.","section":"Section 4, Corollary 4.8"},{"comment":"The sentence 'Conversely, this is the only restraint' identifies the image of Ψ with the intersection of the positive orthant with the hyperplanes A_{jki}+A_{kij} = A_{j'k'i'}+A_{k'i'j'} for interior edges. This identification is load-bearing: it underlies the manifold structure in Proposition 4.9, the completeness proof in Theorem 7.6, and the Finsler results of Section 4.1, but the converse is not proved. Please add the proof: given positive A_{ijk} satisfying the equalities, define l_i := A_{jki}+A_{kij} for any face containing edge i; the equalities make l_i independent of the chosen face, positivity of the A-coordinates gives the triangle inequalities in each face, and gluing the resulting Euclidean triangles along edges yields a Euclidean structure on (S,T).","section":"Section 4, after Proposition 4.3"},{"comment":"The proof that (X^*, Δ) has the convergence-symmetry property is not valid as written. The argument applies the convergence-symmetry hypothesis of X to the double-indexed pairs (p_{n(k),m(k)}) and (q_{n(k),m(k)}) without ensuring that the selected diagonal subsequence satisfies d(p_{n(k),m(k)}, q_{n(k),m(k)}) → 0; the assumption lim_n lim_m d(p_{n,m}, q_{n,m}) = 0 does not by itself imply this for the subsequence chosen from the condition on d(q_{n,m}, p_{n,m}). A correct proof can be obtained by choosing, for each k, representatives and indices m_k, m'_k large enough that d(p_{n_k,m'_k}, q_{n_k,m'_k}) < 1/k and such that the Cauchy-diagonal terms d(p_{n_k,m'_k}, p_{n_k,m_k}) and d(q_{n_k,m'_k}, q_{n_k,m_k}) are small, then applying the convergence-symmetry property of X and a triangle-inequality argument. Please rewrite this step.","section":"Section 5, step (5) in the completion construction"}],"minor_comments":[{"comment":"The asymmetry check uses the points (1,1,1) and (√3/2, √3/2, 1−√3/2), which do not lie in T1; please either normalize them to unit area or note that asymmetry on (R_+^*)^3 descends to T1 by the scaling formula exp(η(λX, λ'Y)) = (λ'/λ) exp(η(X,Y)).","section":"Proposition 2.1"},{"comment":"The Abstract states that the metric is 'a restriction of a non-symmetric version of the classical Thompson distance' and that the paper builds a bridge to Thompson's metrics, but the body does not discuss Thompson distances; please add a remark or soften the claim.","section":"Abstract and body"},{"comment":"The formulations conflate the metrics: problem (4) refers to 'ηm and ηm' and problem (5) to 'ηa and ηa'; presumably these should compare ηm with ηa, and the inequalities should be stated between the two different metrics.","section":"Section 8, open problems (4) and (5)"},{"comment":"There are several typos: 'Cleary' before Definition 4.4, 'W need' in the proof of Lemma 2.13, 'a convex Euclidean structures' in the title of Section 4, 'l_{FP}' in Section 8 (should presumably be l_{Fm}), and in Example 7.7 the sentence after defining Q'_n says 'the area of Qn is also equal to 1' where Q'_n is meant.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern about Corollary 4.8 is accurate and should be addressed before acceptance; the same issue affects the proofs of Proposition 4.10 and Theorem 4.13. The problems are localized and repairable (the affine-path construction works), so I do not recommend rejection. The paper relies on the authors' unpublished preprint [21] for a nontrivial theorem (Theorems 2.16 and 4.13); the editor may wish to verify its status. The single-triangle part of the paper is solid and the counterexample in Example 7.7 is a genuinely useful addition."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the single-triangle part of this paper is solid and worth having; the triangulated-surface part has a genuine gap in one of its main proofs, though the result looks true with a simple fix.\n\nThe metric itself is not new—as the authors acknowledge, it is the Thompson/Funk distance on the simplex. What is new and good: the geodesic criterion (a path is a geodesic iff one coordinate's log-growth dominates), the bigeodesic statement, the Finsler realization F(A,v)=max v_i/A_i, and the completeness theorem for T1. I hand-checked these; they are correct and cleanly proved. The box-Lipschitz reinterpretation (Prop 2.2-2.3) is a nice Thurston analogue, and the completeness formalism in Sections 5-7 is reusable. The forward-incomplete two-triangle example (7.7) is a useful negative result.\n\nThe stress-test concern is real. Corollary 4.8 constructs a bigeodesic in E(S,T)_1 by coordinate-wise geometric interpolation alpha(t)=(A^{1-t}B^t). But E(S,T) is cut out by linear equalities A_{jki}+A_{kij}=A_{j'k'i'}+A_{k'i'j'} per interior edge, and geometric interpolation does not preserve sums; the path generally leaves E(S,T). The fix is easy: use the affine path (1-t)A+tB, since the coordinate maximizing B_i/A_i has the largest logarithmic derivative at every t, and the minimizing coordinate gives the reverse geodesic. So the theorem is likely true but the proof as written is invalid. Since Cor 4.8 is the only support for Theorem 4.13 (via the unpublished [21]), Section 4's geodesic claims currently rest on a broken argument.\n\nOther soft spots: the 'only restraint' claim after Prop 4.3 is asserted without proof—true and elementary, but load-bearing for Sections 4-7. The displayed partial derivatives of G in Section 3 are off by a factor of 2, and the error propagates into the explicit F* formula and the unit-ball vertices in Q. Theorems 2.16 and 4.13 defer to the authors' preprint [21]. Section 8 defines eta_m twice (as a max over triangulations, and as an infimum of F^m-length) without proving the two agree. The asymmetry example in Prop 2.1 uses points outside T1; replacing them with unit-area points fixes it.\n\nNone of this damages the central triangle results. This deserves a serious referee: whoever referees should require the affine-path repair for Cor 4.8, an elementary proof of the parametrization converse, corrected derivatives in Section 3, and proofs rather than citations for 2.16/4.13. With those in place I would accept.","headline":"Single-triangle results are correct and citable; the triangulated-surface geodesic proof has a real gap (fix via affine interpolation), plus a factor-2 derivative typo in Section 3.","tokens_in":22266,"tokens_out":5712,"would_cite":true,"duration_ms":50322,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["32G15","53C70","51K05","51K10","53B40","53C60"],"pacs":[],"model":"deepseek-v4-flash","headline":"An asymmetric log-ratio metric on unit-area Euclidean triangles is complete and Finsler, with geodesics governed by one dominant coordinate.","keywords":["asymmetric metric","Euclidean triangles","singular Euclidean structures","Finsler structure","geodesics","best Lipschitz map","Teichmüller theory","convex polygons"],"falsifier":"For a fixed triangulation of a compact surface with at least one interior edge, enumerate all positive assignments $A_{ijk}$ satisfying the one-linear-equality-per-interior-edge condition and try to build the corresponding Euclidean surface by gluing the triangles; exhibit one assignment that satisfies every equality but cannot be realized by any gluing. Finding such an assignment would disprove the 'only restraint' assertion on which Sections 4-7 rest.","tokens_in":20827,"feed_emoji":"📐","tokens_out":10524,"duration_ms":92934,"temperature":0.7,"pith_summary":"This paper introduces an asymmetric distance $\\eta$ on the space of unit-area marked Euclidean triangles: writing the three edge lengths through semiperimeter-defect coordinates $A_1,A_2,A_3$, the distance from $X$ to $Y$ is $\\log \\max_i\\{A'_i/A_i\\}$. It proves that this metric is Finsler, with infinitesimal norm $\\max_i v_i/A_i$, that geodesics are exactly the paths on which one coordinate's logarithmic growth dominates the other two on every subinterval, and that the space is complete. The same formula defines an asymmetric metric on the space $E(S,T)_1$ of unit-area singular Euclidean structures on a compact surface with a fixed triangulation, where it is again Finsler and bigeodesic, and complete after arithmetic or max symmetrisation. A two-triangle example shows the asymmetric metric on triangulated surfaces need not have the convergence-symmetry property and can be forward incomplete. The metric is, in the abstract's phrase, a restriction of a non-symmetric version of the classical Thompson distance, and the paper presents it as a Euclidean analogue of the classical asymmetric metric on Teichmüller space.","feed_headline":"Ratio metric makes triangle space complete and Finsler","feed_subtitle":"A log-ratio distance is Finsler and complete on triangles, extends to fixed triangulations, and fails there in a two-triangle example","key_machinery":"The central object is the change of coordinates $A_i=(l_j+l_k-l_i)/2$ on the edge lengths of a Euclidean triangle, the semiperimeter-defect coordinates, in which Heron's formula becomes $\\mathrm{area}^2=(A_1+A_2+A_3)A_1A_2A_3$ and the unit-area condition is a smooth hypersurface. The metric compares two triangles by the logarithm of the maximum ratio of these coordinates, which is also the logarithm of the best Lipschitz constant of a label-preserving map between boxes in $\\mathbb{R}^3$ with side lengths $A_i$. The Finsler norm is $F(A,v)=\\max_i v_i/A_i$, whose unit ball in the equivalent quadrant model is a right triangle. The geodesic criterion, one coordinate's logarithmic growth dominating the others on every subinterval, makes bigeodesics explicit: paths of the form $A_i(t)=A_i^{1-t}(A'_i)^t$, rescaled to unit area, work for any pair of endpoints. For triangulated surfaces the same coordinates are taken face by face, and the image of the parametrization is asserted to be cut out by exactly one linear equality per interior edge.","core_discovery":"The paper's central claim is that $\\eta(X,Y)=\\log \\max_i\\{A'_i/A_i\\}$ on the unit-area hypersurface $T_1=\\{(A_1,A_2,A_3):A_i>0,\\ (A_1+A_2+A_3)A_1A_2A_3=1\\}$ is a complete asymmetric Finsler metric. A path is a geodesic precisely when, after rescaling each point to unit area, some face coordinate $A_j$ has logarithmic growth at least as large as the other two coordinates on every subinterval; any two points are joined by a bigeodesic. The infinitesimal norm is $F(A,v)=\\max_i v_i/A_i$, and the same formula defines a Finsler metric on the unit-area slice $E(S,T)_1$ of Euclidean structures on a fixed triangulation. The paper proves $T_1$ is complete and has the convergence-symmetry property; for $E(S,T)_1$, completeness holds for the symmetrised metrics, while the asymmetric metric itself can fail: a disc triangulated by two triangles admits a forward-Cauchy sequence of unit-area quadrilaterals that degenerates to a segment. Along the way the paper develops a completion theory for asymmetric metric spaces based on forward and backward Cauchy sequences and the convergence-symmetry property.","pith_inferences":["If the 'only restraint' premise holds, then $E(S,T)_1$ is a linear section of a positive orthant, so the difference between complete and incomplete examples should be governed by which boundary directions of that cone are reachable by forward-Cauchy sequences; classifying triangulations by this boundary combinatorics is a natural next step.","Example 7.7 suggests the completion of $E(S,T)_1$ for $\\eta(T)$ adds flat structures in which some face coordinates collapse to zero, with forward and backward convergence distinguishing the two sides of the collapse; computing the metric completion for the two-triangle disc would test this.","The box-Lipschitz reformulation indicates that $\\eta$ can be compared with other Thompson-type metrics on cones of labelled convex bodies; if that comparison holds, the geodesic and Finsler formulas should transfer to spaces of labelled polytopes with the same coordinate ratios."],"forward_implications":["Between any two unit-area marked triangles there is a bigeodesic, and the geodesic condition is explicit: one coordinate's logarithmic growth dominates the others on every subinterval.","$T_1$ is complete as a metric space and has the convergence-symmetry property, so every Cauchy sequence converges and no nontrivial completion is needed there.","For every $t\\in(0,1)$, the space $E(S,T)_1$ is complete with respect to the max-symmetrised metric $\\eta(T)^m_t$, hence also with respect to the arithmetic symmetrisation $\\eta(T)^a_t$.","The formulas push forward to the space of unit-area convex Euclidean polygons with $n$ distinguished boundary points, where they define Finsler metrics $\\eta_m$ and $\\eta_a$; the paper leaves geodesics and completeness of these metrics open.","The two-triangle disc example separates the asymmetric theory from its symmetrisation: the symmetrised space is complete while the asymmetric metric is forward incomplete."],"supporting_citations":[{"why":"Supplies the length-ratio and best-Lipschitz equivalence that the box model of $\\eta$ mirrors.","marker":"[23]"},{"why":"Provides the earlier Euclidean acute-triangle metric and minimal-stretch-map results that $\\eta$ is contrasted with.","marker":"[19]"},{"why":"Studied the acute-triangle space whose symmetric metric is incomplete, motivating the completeness questions here.","marker":"[14]"},{"why":"Supplies Theorem 5.1 used to conclude the arithmetic symmetrisations $\\eta_t^a$ are Finsler with the stated norm.","marker":"[21]"},{"why":"Supplies the definitions of geodesic, weak norm, and weak Finsler structure used in the paper.","marker":"[17]"},{"why":"Provides the example of a degenerate weak norm on Euclidean tori that motivates the weak-norm definition.","marker":"[3]"}],"fun_headline_variants":["Asymmetric Finsler metric completes triangle space","Euclidean triangles get complete asymmetric Finsler metric","Log-ratio distance yields complete Finsler metric on triangles","Triangle space complete under non-symmetric Thompson metric","Finsler bridge between Thompson and Thurston on triangles"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise for the triangulated-surface results is the assertion after Proposition 4.3 that the image of the face-coordinate parametrization is cut out exactly by one linear equality per interior edge; if hidden constraints existed, $E(S,T)_1$ and the metric $\\eta(T)$ would be different objects and the geodesic and Finsler theorems would not follow.","fun_headline_variants_meta":{"raw":{"variants":["Asymmetric Finsler metric completes triangle space","Euclidean triangles get complete asymmetric Finsler metric","Log-ratio distance yields complete Finsler metric on triangles","Triangle space complete under non-symmetric Thompson metric","Finsler bridge between Thompson and Thurston on triangles"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000568,"raw_usage":{"total_tokens":2727,"prompt_tokens":1017,"completion_tokens":1710,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":633,"completion_tokens_details":{"reasoning_tokens":1632}},"tokens_in":633,"tokens_out":1710,"duration_ms":15725,"temperature":1.0,"reasoning_tokens":1632,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T10:46:45.559835+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a fixed triangulation of a compact surface with at least one interior edge, enumerate all positive assignments $A_{ijk}$ satisfying the one-linear-equality-per-interior-edge condition and try to build the corresponding Euclidean surface by gluing the triangles; exhibit one assignment that satisfies every equality but cannot be realized by any gluing. Finding such an assignment would disprove the 'only restraint' assertion on which Sections 4-7 rest.","supporting_citations":[{"cited_title":"Thurston: Minimal stretch maps between hyperbolic surfaces","cited_arxiv_id":null,"evidence_quote":"Supplies the length-ratio and best-Lipschitz equivalence that the box model of $\\eta$ mirrors."},{"cited_title":"Sa˘ glam and A","cited_arxiv_id":null,"evidence_quote":"Provides the earlier Euclidean acute-triangle metric and minimal-stretch-map results that $\\eta$ is contrasted with."},{"cited_title":"Miyachi, K","cited_arxiv_id":null,"evidence_quote":"Studied the acute-triangle space whose symmetric metric is incomplete, motivating the completeness questions here."},{"cited_title":"Sa˘ glam, K","cited_arxiv_id":null,"evidence_quote":"Supplies Theorem 5.1 used to conclude the arithmetic symmetrisations $\\eta_t^a$ are Finsler with the stated norm."},{"cited_title":"Papadopoulos and M","cited_arxiv_id":null,"evidence_quote":"Supplies the definitions of geodesic, weak norm, and weak Finsler structure used in the paper."},{"cited_title":"Belkhirat, A","cited_arxiv_id":null,"evidence_quote":"Provides the example of a degenerate weak norm on Euclidean tori that motivates the weak-norm definition."}],"review_version":1}