{"id":"85cb69ba-3e30-4c1d-9879-550f80446b23","arxiv_id":"2506.03749","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For Finsler metrics with bi-geodesics, the arithmetic and maximum symmetrisations of the distance are shown to be Finsler, with explicit Lagrangians, generalising known Hilbert and Funk results.","lead":"This paper studies two natural ways to blend a Finsler metric with its own reverse, arithmetic and maximum symmetrisation, and identifies when the resulting distance functions are themselves Finsler metrics. It applies the framework to Funk, Hilbert, and Teichmüller geometries, showing that weighted arithmetic Funk metrics are Finsler with straight-line geodesics.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Corollary 6.5 overreaches for unbounded convex domains: the half-space example has a degenerate Lagrangian, violating Definition 4.1(3).","rationale":"I read Theorem 5.3's proof carefully and it appears sound under its explicit bi-geodesic hypothesis; the inequality chain in Eq. (11) is valid, and the hypothesis is indeed satisfied for bounded Funk domains, where straight segments are bi-geodesics. The reader's weakest assumption focuses on the bi-geodesic condition and its unverified status for Teichmüller spaces. That is a legitimate scope concern, but it does not expose a flaw in the proved theorem: the condition is stated as a hypothesis, and Section 7 presents open questions rather than claiming new Teichmüller results. The more load-bearing concrete problem is in Corollary 6.5's unqualified statement for all open convex sets. Section 6's setup explicitly allows unbounded domains, and the paper's own half-space example yields a Lagrangian that vanishes on horizontal vectors, violating the paper's Definition 4.1(3) and making the induced distance non-positive on horizontal lines. This means the headline application to Funk metrics overreaches unless boundedness is imposed or the definition of Finsler structure is weakened to allow degenerate seminorms. Since the fix is a straightforward restriction and the general framework is unaffected, the reader's CONDITIONAL verdict remains appropriate; the condition should include boundedness of Ω in the Funk/Finsler corollaries.","tokens_in":23341,"tokens_out":18012,"duration_ms":172079,"concrete_test":"Recompute the upper half-space example in Section 6 from the definition p(x,v)=inf{t>0 | x+v/t∈Ω}. Take x=(0,…,0,1) and v=(1,0,…,0). Verify p(x,v)=p(x,−v)=0, so p^a_t(x,v)=0 for all t∈[0,1], violating Def 4.1(3). Then test Corollary 6.5 on x=(0,…,0,1) and y=(1,0,…,0,1): the Funk distances vanish in both directions, giving F^a_t(x,y)=0 for x≠y, contradicting the claim that F^a_t is a Finsler metric. If Ω is intended to be bounded, repeat the computation on the unit ball to confirm the corollary holds there.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim in the Funk/Hilbert application is Corollary 6.5: for every open convex set Ω, the arithmetic weighted Funk metric F^a_t is Finsler with Lagrangian p^a_t. But Section 6 allows Ω to be any closed convex set with nonempty interior, and the paper's own upper half-space example H={x_n>0} gives p(x,v)=0 for horizontal vectors. Indeed, from p(x,v)=inf{t>0 | x+v/t∈Ω}, if v_n>0 the ray x+v/t stays in H for all t>0, so p=0; if v_n=0, p=0 as well. The displayed p(x,v)=max(v_n/x_n,0) appears to have the sign reversed, but either way horizontal vectors have p=0. Therefore p^a_t(x,v)=(1−t)p(x,v)+tp(x,−v) also vanishes on horizontal vectors, contradicting Definition 4.1(3), which requires F(x,v)=0 iff v=0. The induced weak metric is not positive: two points on the same horizontal level have F(x,y)=F(y,x)=0, hence F^a_t(x,y)=0 for x≠y. Thus Corollary 6.5 is false as stated unless Ω is bounded (or the Finsler definition is relaxed to allow degenerate seminorms, which the paper does not do). This is load-bearing because it is the advertised application to Funk geometry, and the same issue affects the max-Funk discussion in Remark 6.3.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies two one-parameter families of metrics derived from a Finsler metric F: the arithmetic family d(F)^a_t = (1-t)d(F)+t d(F)^{-1} and the max family d(F)^m_t = max{(1-t)d(F), t d(F)^{-1}}. It gives conditions under which these distance functions are induced by Finsler structures with Lagrangians F^a_t = (1-t)F+tF(x,-v) and F^m_t = max{(1-t)F, tF(x,-v)}. The main positive results are Theorem 5.1 (characterization via common geodesics for two Finsler metrics), Theorem 5.3 (bi-geodesic hypothesis implies the arithmetic symmetrization is Finsler), Theorem 5.4 (sign condition for the max Lagrangian), and a necessary-and-sufficient condition in Theorem 5.5. The paper also develops weak-metric tools (completeness, Arzelà–Ascoli, Hopf–Rinow in the asymmetric setting) and applies the results to Funk and Hilbert geometries, where the arithmetic weighted Funk metrics are shown to have straight-line geodesics. Section 7 formulates open questions for Thurston-like metrics on Teichmüller spaces.","tokens_in":23607,"tokens_out":11140,"duration_ms":104961,"significance":"If the main theorems are correct, they provide a clean synthesis of known symmetrization phenomena in Finsler and Funk/Hilbert geometry, with machine-checkable-level detail in the proofs of Theorems 5.1, 5.3, and 5.4. The explicit use of absolute continuity and bi-geodesics is a genuine improvement over earlier heuristic treatments. The application to weighted Funk metrics and the connection to Hilbert's fourth problem are attractive. However, the advertised application in Corollary 6.5 is not valid for unbounded convex domains, and the necessity proof of Theorem 5.5 has a gap; these need to be resolved before the results can be accepted in full.","major_comments":[{"comment":"The statement of Corollary 6.5 is false as written for unbounded convex sets. For the upper half-space H={x_n>0}, the Funk Lagrangian p(x,v)=inf{t>0 | x+v/t∈H} vanishes whenever v_n≥0, including all horizontal vectors v with v_n=0. Consequently p^a_t(x,v)=(1-t)p(x,v)+tp(x,-v) also vanishes on horizontal vectors, contradicting Definition 4.1(3), which requires F(x,v)=0 if and only if v=0. The induced distance satisfies F^a_t(x,y)=0 for two distinct points on the same horizontal level, so p^a_t is not a Finsler structure in the sense of the paper. The displayed formula p(x,v)=max(v_n/x_n,0) is also incorrect: the correct expression is p(x,v)=max(-v_n/x_n,0), and the subsequent piecewise formula for p^a_t is consistent with this corrected p. The corollary can be repaired by restricting to bounded convex sets (or more generally to convex sets containing no affine line), but this restriction is not stated anywhere in Section 6; indeed the section explicitly allows arbitrary closed convex sets and presents the upper half-space as an example. This affects the advertised application to Funk geometry and the Hilbert-fourth-problem interpretation in Proposition 6.1.","section":"Section 6, Corollary 6.5 and the upper half-space example"}],"minor_comments":[{"comment":"The notation 'length dm (α[a,t0])' is slightly ambiguous; it would be clearer to write 'the length of the subarc α|[a,t0] with respect to the norm ‖·‖_m'.","section":"Section 2, Example 2.1"},{"comment":"The metric d(F) is denoted dF in Definition 4.14 and Lemma 4.15 without being defined; please use d(F) consistently throughout.","section":"Section 4, Definition 4.14 and Lemma 4.15"},{"comment":"The sentence 'She other inequality' contains a typo; it should be 'The other inequality'.","section":"Section 4, proof of Proposition 4.13"},{"comment":"There are several typos in the displayed formulas: the final line writes 'd(max{(1-t)F1,F2})' where tF2 is meant, and earlier 'max{td(F1),(1-t)d(F2)}' swaps the arguments. These make the case analysis harder to follow.","section":"Section 5, Theorem 5.5 proof"},{"comment":"The section begins by taking Ω to be a closed convex set, but Corollary 6.5 refers to 'an open convex set'; the relationship between these two conventions (and the domain on which the Funk metric is defined) should be clarified.","section":"Section 6, first paragraph"},{"comment":"The outline says Busemann's axiom is introduced in Section 4, but it is actually defined in Definition 3.2 in Section 3; please correct the cross-reference.","section":"Section 1, outline"}],"recommendation":"major_revision","confidential_remarks":"The paper contains useful general results and the exposition is largely careful, but the unbounded convex domain issue in Section 6 is a genuine mathematical error that affects a headline application. The authors should either restrict the domain to bounded convex sets or explicitly extend Definition 4.1 to allow degenerate seminorms and then adjust the statements and proofs. The proof of Theorem 5.5 also needs a substantial repair. These are fixable within the scope of the paper, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The core of this paper is worth your time. Theorems 5.1, 5.3, 5.4 and 5.5 give clean necessary-and-sufficient conditions for arithmetic and max symmetrizations of Finsler distances to be Finsler, with explicit Lagrangians. These are real results that generalize known facts about Hilbert and Funk metrics in a useful way, and the proofs look essentially correct. The Hopf–Rinow variant and the completeness results in Section 4 are also nice additions, even if some proofs are sketched or delegated.\n\nBut the paper has a genuine flaw in Section 6, exactly where the advertised application lives. Corollary 6.5 claims that the arithmetic weighted Funk metric on every open convex set is Finsler with Lagrangian p^a_t. That is false for unbounded domains. The half-space example makes it concrete: for horizontal vectors, p(x,v)=p(x,-v)=0, so p^a_t vanishes on nonzero vectors, violating Definition 4.1(3). The displayed formula p(x,v)=max(v_n/x_n,0) is also just wrong; the correct expression is max(-v_n/x_n,0). This is not a cosmetic issue: the induced weak metric is not positive on points sharing the same horizontal level, so it cannot be Finsler in the paper's own sense. The fix is to restrict to bounded convex domains, or explicitly relax the Finsler definition to allow degenerate seminorms, and then adjust Proposition 6.1 accordingly.\n\nThe other soft spots are less severe. The bi-geodesic hypothesis in Theorem 5.3 is strong, and Section 7 does not verify it for Thurston-like metrics on Teichmüller space; the abstract over-promises what the paper delivers there. The proof of Theorem 5.5's necessity is hard to follow and has typos, and Prop 6.2 is delegated to the literature. These are presentation issues, not fatal.\n\nThe reader's verdict was CONDITIONAL with moderate confidence; I think that is about right, but the unbounded-domain problem means the condition is not just a gap but a real error in a corollary. Still, the main general framework is sound and deserves a serious referee. I would send it to peer review, expecting heavy revision.\n\nReading group: yes, I'd bring it up if only to discuss the main theorems and why the Funk application collapses on unbounded sets.","headline":"Solid general theorems on symmetrizing Finsler metrics, but the advertised Funk-metric application is spoiled by a genuine degeneracy issue on unbounded domains.","tokens_in":24214,"tokens_out":4750,"would_cite":true,"duration_ms":50588,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["53C70","51K05","51K10","53B40","53C60"],"pacs":[],"model":"deepseek-v4-flash","headline":"Under a bi-geodesic condition, arithmetic symmetrisation of any Finsler metric is again Finsler, with the weighted norm as Lagrangian.","keywords":["Symmetrisation of a Finsler metric","Hilbert metric","weighted Funk metric","families of Finsler metrics","complete non-symmetric metric","Teichmüller space","Teichmüller spaces of Euclidean surfaces","Thurston metric"],"falsifier":"Integrate the Lagrangian $p^a_t$ given in Section 6 for the upper half-space model, namely $p^a_t(x,v)=t v_n/x_n$ for $v_n>0$ and $(1-t)|v_n|/x_n$ for $v_n<0$, along the straight segment between two points and compare with the weighted Funk distance $F^a_t$; any $t\\in(0,1)$ for which the two numbers differ would refute Corollary 6.5 and the bi-geodesic transfer principle behind it.","tokens_in":23094,"feed_emoji":"📐","tokens_out":12005,"duration_ms":112903,"temperature":0.7,"pith_summary":"A Finsler metric is a continuous family of possibly asymmetric norms on the tangent spaces of a manifold, and it induces a distance that need not be symmetric. This paper asks when weighted combinations of such a distance with its reverse, namely $d(F)^a_t(x,y)=(1-t)d(F)(x,y)+t d(F)(y,x)$ and the max version, are again induced by a Finsler norm. The central answer is that the arithmetic combination is Finsler whenever every pair of points is joined by a bi-geodesic, a path that is length-minimizing in both directions simultaneously, and the new Lagrangian is simply $F^a_t(x,v)=(1-t)F(x,v)+tF(x,-v)$. In the Funk geometry of convex domains, straight segments are bi-geodesics, so the weighted arithmetic Funk metrics are Finsler with straight-line geodesics, giving a one-parameter family of solutions to Hilbert's fourth problem. The same framework yields criteria for weighted sums and maxima of two Finsler metrics to be Finsler and applies to asymmetric metrics on Teichmüller spaces.","feed_headline":"Symmetrized Finsler metrics stay Finsler when geodesics run both ways","feed_subtitle":"Weighted sums of a metric and its reverse stay Finsler; weighted Funk metrics gain straight geodesics.","key_machinery":"The load-bearing object is the bi-geodesic: an absolutely continuous path $\\gamma$ with $\\ell_F(\\gamma)=d(F)(x,y)$ and $\\ell_F(\\gamma^{-1})=d(F)(y,x)$. The proof of Theorem 5.3 lets this single path attain the two infima in Lemma 3.7 simultaneously, converting the general inequality $\\delta(\\ell(F^a_t))\\ge d(F)^a_t$ into equality. The companion machinery is the identification of variational length with integral Finsler length for absolutely continuous paths (Lemma 4.7) and the identity $(\\ell_F)_t=\\ell_{F^a_t}$ (Corollary 4.20), which pass the arithmetic combination through the length-to-distance construction. For the max family, the controlling condition is a sign condition on the difference $(1-t)F_1-tF_2$ along a common geodesic.","core_discovery":"The core discovery is that reversibility of geodesics, not symmetry of the norm, is the controlling condition for whether symmetrised distances stay Finsler. Theorem 5.3 states that if a Finsler structure $F$ on a smooth manifold admits an absolutely continuous bi-geodesic between every pair of points, then for every $t\\in[0,1]$ the arithmetic weighted distance $d(F)^a_t$ is induced by the Finsler structure $F^a_t(x,v)=(1-t)F(x,v)+tF(x,-v)$, and every bi-geodesic of $F$ is also a bi-geodesic of $F^a_t$. Because straight segments are bi-geodesics for the Funk metric on a convex domain, Corollary 6.5 makes the weighted arithmetic Funk metrics Finsler with straight-line geodesics. The paper further shows that a weighted sum $(1-t)F_1+tF_2$ induces the weighted distance exactly when a common absolutely continuous geodesic exists, and that the max version requires, in the smooth complete case, a common geodesic along which one Lagrangian dominates the other.","pith_inferences":["The bi-geodesic hypothesis suggests a concrete check for other asymmetric metrics: if some pair lacks a two-way shortest path, the arithmetic symmetrisation could still be Finsler, but the equality in Theorem 5.3 would not follow from this argument; Section 7's Teichmüller metrics are the natural place to test whether the condition is necessary.","The sign condition in the max theorem looks like a no-overtaking condition along a common geodesic; on convex domains, testing it for the weighted max Funk metrics would explain exactly where the counterexample in Remark 6.3 enters and whether other weights behave better.","Using the explicit Lagrangian in the ball and upper half-space models, one could compute flag curvature or projectivity of the weighted Funk metrics; the paper establishes geodesics and Finslerity but leaves those derived quantities open."],"forward_implications":["For any convex domain, the weighted arithmetic Funk metrics $F^a_t$ are Finsler for every $t$, and the straight Euclidean segments are their geodesics, so this is a family of solutions to Hilbert's fourth problem.","The Hilbert metric is the $t=1/2$ member of this family, and its Finsler Lagrangian is recovered as the arithmetic symmetrisation of the Funk Lagrangian, now extended from $t=1/2$ to all weights.","Any Finsler manifold satisfying the bi-geodesic hypothesis inherits a geodesic-sharing property: the same paths that are bi-geodesics for $F$ are bi-geodesics for $F^a_t$.","Sums and maxima of complete Finsler structures are complete, and forward and backward convergence always agree for Finsler-induced distances, which reproduces the convergence behaviour known for Thurston's and the earthquake metrics.","For $t\\neq 0$, the weighted arithmetic Funk metric is uniquely geodesic precisely when the boundary contains no pair of non-collinear Euclidean segments in a common plane, matching the Hilbert-metric criterion."],"supporting_citations":[{"why":"It supplies the infinitesimal Finsler form of the Funk metric and of the Hilbert metric that Corollary 6.5 turns into the weighted Lagrangian $p^a_t$.","marker":"[34]"},{"why":"It supplies the earlier description of the Minkowski functional of the arithmetic symmetrisation of a Finsler metric, which Theorem 5.3 generalises.","marker":"[26]"},{"why":"It supplies the classical result that variational length equals integral Finsler length for piecewise $C^1$ paths, used as Lemma 4.6.","marker":"[6]"},{"why":"It extends the length identity to absolutely continuous paths, which Lemma 4.7 needs for the bi-geodesic hypothesis in Theorem 5.3.","marker":"[35]"},{"why":"It supplies the asymmetric compactness theorem used to prove the Hopf–Rinow analogue that yields geodesics in Proposition 4.13.","marker":"[8]"},{"why":"It supplies the smooth-geodesic existence theorem for complete $C^\\infty$-Finsler structures used in the necessity direction of Theorem 5.5.","marker":"[2]"}],"fun_headline_variants":["Bi-geodesics settle when symmetrized Finsler metrics stay Finsler","Symmetrized metrics stay Finsler if geodesics are bidirectional","Weighted Funk metrics become Finsler with straight geodesics","Geodesic reversibility is the key for Finsler symmetrization"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that every pair of points can be joined by a path that is shortest in both directions at once, which Theorem 5.3 requires and Section 7 does not verify for the Thurston-like metrics it discusses.","fun_headline_variants_meta":{"raw":{"variants":["Bi-geodesics settle when symmetrized Finsler metrics stay Finsler","Symmetrized metrics stay Finsler if geodesics are bidirectional","Weighted Funk metrics become Finsler with straight geodesics","Geodesic reversibility is the key for Finsler symmetrization"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000629,"raw_usage":{"total_tokens":2902,"prompt_tokens":938,"completion_tokens":1964,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":554,"completion_tokens_details":{"reasoning_tokens":1883}},"tokens_in":554,"tokens_out":1964,"duration_ms":16507,"temperature":1.0,"reasoning_tokens":1883,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T10:56:43.599944+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Integrate the Lagrangian $p^a_t$ given in Section 6 for the upper half-space model, namely $p^a_t(x,v)=t v_n/x_n$ for $v_n>0$ and $(1-t)|v_n|/x_n$ for $v_n<0$, along the straight segment between two points and compare with the weighted Funk distance $F^a_t$; any $t\\in(0,1)$ for which the two numbers differ would refute Corollary 6.5 and the bi-geodesic transfer principle behind it.","supporting_citations":[{"cited_title":"Troyanov, Funk and Hilbert geometries from the Finsl erian view- point, In: Handbook of Hilbert geometry (A","cited_arxiv_id":null,"evidence_quote":"It supplies the infinitesimal Finsler form of the Funk metric and of the Hilbert metric that Corollary 6.5 turns into the weighted Lagrangian $p^a_t$."},{"cited_title":"Papadopoulos and M","cited_arxiv_id":null,"evidence_quote":"It supplies the earlier description of the Minkowski functional of the arithmetic symmetrisation of a Finsler metric, which Theorem 5.3 generalises."},{"cited_title":"Busemann and W","cited_arxiv_id":null,"evidence_quote":"It supplies the classical result that variational length equals integral Finsler length for piecewise $C^1$ paths, used as Lemma 4.6."},{"cited_title":"Zhang & W","cited_arxiv_id":null,"evidence_quote":"It extends the length identity to absolutely continuous paths, which Lemma 4.7 needs for the bi-geodesic hypothesis in Theorem 5.3."},{"cited_title":"Collins and J","cited_arxiv_id":null,"evidence_quote":"It supplies the asymmetric compactness theorem used to prove the Hopf–Rinow analogue that yields geodesics in Proposition 4.13."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It supplies the smooth-geodesic existence theorem for complete $C^\\infty$-Finsler structures used in the necessity direction of Theorem 5.5."}],"review_version":1}