{"id":"2ee29e42-0a95-4828-adb1-c09e990b8dcf","arxiv_id":"2507.22737","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"In globally hyperbolic spacetimes, the locus of multiple maximizing geodesics is locally contractible and homotopy equivalent to the causal future minus the Lorentzian Aubry set.","lead":"This paper proves that, in any globally hyperbolic spacetime, the set of pairs of events connected by more than one fastest path is a topologically tame object (locally contractible). It also shows that this set has the same shape as the causal future with a newly defined 'Aubry set' removed, a structural result already known on Riemannian manifolds and now extended to Lorentzian geometry.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 5.6 is the load-bearing import: the branch 'two maximizing geodesics ⇒ no maximizing extension' is only cited, with the paper itself noting omitted proofs in [1]; if this branch fails, Props 5.16, 5.26, 5.28 and the main theorems collapse.","rationale":"The Reader's weakest_assumption identifies exactly the same load-bearing premise: Theorem 5.6, the cut-locus classification, is imported rather than proved, and the paper explicitly flags that parts of its cited source omit proofs. I agree this is the single most structural risk. The semiconvexity/semiconcavity estimates in Sections 3 and 4 are also difficult to verify from the text, but they are technical tools; Theorem 5.6 is the conceptual pivot through which every main topological consequence passes. In particular, the deformation retraction in Prop 5.26, the homotopy inverse in Prop 5.28, and even the timelike local-contractibility argument all rely on the statement that a second distinct maximizing geodesic forces the original one to stop maximizing. The paper's own note that the proof is omitted in [1] is an explicit admission of missing support, so this is not a manufactured concern. The proposed test, a direct proof of the branching-to-no-extension implication via the corner argument, would settle whether the concern is merely cosmetic or actually invalidates the argument. I found no internal inconsistency and no reason to change the Reader's verdict: the paper should remain CONDITIONAL pending that verification. If the test succeeds, the paper would be acceptable; if it fails, the homotopy equivalences would need substantial reworking.","tokens_in":44568,"tokens_out":14646,"duration_ms":181546,"concrete_test":"Attempt a fully self-contained proof of Theorem 5.6(ii): let γ,η be distinct maximizing geodesics from x to y, and suppose γ extends to a maximizing geodesic on [0,1+ε]. Show the concatenation η followed by γ|[1,1+ε] is maximizing from x to γ(1+ε), using d(x,y)+d(y,z) ≤ d(x,z) and length ≤ d; then conclude it is a pregeodesic by [18] Thm 2.9, contradicting the corner unless η and γ coincide as unparametrized geodesics. If this proof succeeds, the cited gap is closed and the central argument is sound; if it requires extra hypotheses, such as timelike-only or geodesic completeness, then Props 5.16, 5.26 and 5.28 need repair and the conditional verdict stands.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central chain is: local contractibility of NU and the homotopy equivalences NU -> Cut_M -> J^+\\A are obtained by showing NU_t -> Cut^t_M is a homotopy equivalence (Prop 5.28 via Cor 5.12) and Cut_M is a strong deformation retract of J^+\\A (Prop 5.26). Both arguments repeatedly invoke Theorem 5.6, not as an auxiliary fact but as the mechanism identifying cut points with conjugate points or non-unique maximizers. In particular, Lemmas 5.15, 5.20 and 5.25 use the assertion that if a second distinct maximizing geodesic to y exists, then a given maximizing geodesic ceases to be maximizing beyond y. The paper's proof of this direction says it 'essentially follows' from [18] Thm 2.9 after [1] Cor 9.4 and 9.11, 'although the proofs are omitted there.' [18] Thm 2.9 states that maximizing causal curves are pregeodesics; it does not by itself, as written, contain the branching statement. The missing step is nontrivial: one must prove, e.g. by a corner/concatenation argument plus the reverse triangle inequality, that a maximizing extension beyond y together with the second geodesic would produce a maximizing curve with a corner, contradicting pregeodesicity. If that step cannot be completed, then NU_t ⊆ Cut^t_M, the well-definedness of β, and both deformation retractions lose their foundation. This is the same structural risk identified by the Reader.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper extends the Riemannian results of Cannarsa, Cheng and Fathi on singularities of distance functions to the setting of a globally hyperbolic spacetime. It defines a Lorentzian future Aubry set A(x) and a global Aubry set A, proves that the set NU(M,g) of causally related pairs admitting more than one maximizing geodesic is locally contractible (Theorem 1.2), and proves that the inclusions NU(M,g) → J^+\\A and NU_t(M,g) → I^+\\A are homotopy equivalences, together with fixed-endpoint versions (Theorem 1.5). The main technical work is in Sections 3 and 4, where the paper establishes local semiconvexity and semiconcavity, hence local C^1-regularity, of the Lax-Oleinik semigroup acting on a point-mass initial condition χ_x for the non-Tonelli Lorentzian Lagrangian. Section 5 then uses this regularity to construct homotopies from the cut locus to the non-uniqueness set, and separately shows that the cut locus is a strong deformation retract of J^+\\A and I^+\\A.","tokens_in":44901,"tokens_out":7273,"duration_ms":94475,"significance":"If the results are correct, they are substantial. The paper proves new topological rigidity phenomena for Lorentzian geometry: the non-uniqueness locus of maximizing geodesics and the causal future minus the Aubry set have the same homotopy type, and the lightlike boundary case is handled rather than being excluded. The author works in a setting where the Tonelli hypotheses fail, and the estimates of Sections 3 and 4 are detailed and appear internally coherent. The paper also gives credit to the Riemannian template of Cannarsa–Cheng–Fathi and clearly identifies what is new in the Lorentzian problem. The main risk is not in the semigroup analysis but in a single imported classification theorem about cut points, as detailed below.","major_comments":[{"comment":"Theorem 5.6 is load-bearing, and the branch that a second distinct maximizing geodesic forces the geodesic to cease to be maximizing beyond the cut point is not proved in the paper. The paper says this follows from Corollaries 9.4 and 9.11 of [1], adds that the proofs are omitted there, and refers to [18, Theorem 2.9], which states only that maximizing causal curves are pregeodesics. That statement does not, by itself, contain the branching assertion. The missing step is nontrivial: one must show, for example by a concatenation/corner argument plus the reverse triangle inequality, that if one of the two geodesics admitted a maximizing extension beyond y, a maximizing causal curve with a corner would exist, contradicting pregeodesicity. This implication is used to prove NU_t ⊆ Cut^t_M and to prove that β(x,y) is well-defined; without it, the deformation retractions in Propositions 5.16, 5.26 and 5.28 and the homotopy equivalences of Theorem 1.5 lose their foundation. Please add a complete proof of this implication or give a precise reference in which it is proved in the stated generality.","section":"Theorem 5.6 and its uses in Lemmas 5.15, 5.20, 5.25 and Propositions 5.16, 5.19, 5.26, 5.28"},{"comment":"The well-definedness and continuity of the map β, and hence the strong deformation retraction of J^+\\A onto Cut_M, depend on the unproved branch of Theorem 5.6 in two separate places: first, when two distinct maximizing geodesics from x to y are assumed to imply that y is the cut point along both, and second, when the claim “if (x,y) ∉ A then the maximal extension is not maximizing” is used in Lemma 5.20. Lemma 5.15, Lemma 5.20 and Lemma 5.25 each invoke this same missing classification. The paper should state these dependencies explicitly and either prove the underlying cut-locus classification or isolate it as a clearly marked lemma with a complete proof before using it in the main theorems.","section":"Lemma 5.25 and Proposition 5.26"}],"minor_comments":[{"comment":"There are several typographical errors: “Othwerwise” in Lemma 2.17, “respestively” in the Appendix, “from from” in the Introduction, and “satiyfying” in Theorem 5.11 and Corollary 5.12. These are harmless but should be corrected.","section":"Throughout"},{"comment":"In the proof of Corollary 5.12, the text writes “(U ∩ V)” where the product neighbourhood U × V is meant; this should be corrected.","section":"Corollary 5.12 proof"},{"comment":"In Lemma 5.22, the bump functions ρ_j are introduced as maps M → [0,1], but their arguments are points (x,y) in J^+\\A ⊆ M×M; they should be defined on M×M.","section":"Lemma 5.22 proof"},{"comment":"Reference [1] is cited as “John K Beem. Global Lorentzian Geometry, Second Edition”, but the standard attribution is Beem–Ehrlich–Easley; the author list should be corrected.","section":"Reference [1]"},{"comment":"The symbol for the non-uniqueness set appears as NU, N U, and ⅇ NU in different places; a single notation would improve readability.","section":"Notation"}],"recommendation":"major_revision","confidential_remarks":"The paper is a good fit for a mathematical analysis journal and contains substantial, clearly presented work. My main concern is the dependence of the central homotopy equivalences on a cut-locus classification whose crucial converse direction is only cited, with the author himself noting that the proofs are omitted in the cited source. I would advise requiring a complete proof of that implication, or a genuinely precise reference with a proof, before acceptance. If the missing implication turns out to be false in some globally hyperbolic spacetime, the main theorems of Section 5 would collapse; the rest of the paper appears sound and would still be valuable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The short version: this paper proves something new and substantive. It extends Cannarsa-Cheng-Fathi's local contractibility and homotopy equivalence results to globally hyperbolic spacetimes, introducing a Lorentzian Aubry set and proving the inclusion chain NU(M,g) -> Cut_M -> J^+\\A is a homotopy equivalence. Theorems 1.2 and 1.5 are genuinely new, and the techniques are adapted to the Lorentzian setting rather than imported verbatim.\n\nWhat's good: Sections 3-4 contain intricate semiconvexity/semiconcavity estimates for the Lorentzian Lax-Oleinik semigroup; they look carefully done, and the null-boundary case in the local contractibility proof (Lemmas 5.13-5.14) is a real addition. The strong deformation retracts from Cut_M onto J^+\\A, with speeds controlled by the continuous functions phi_± and beta, are constructed with genuine care. The paper also credits prior work honestly, including its own preprints, and it explicitly flags the one place where the proof is incomplete.\n\nThe main soft spot is Theorem 5.6, the cut-locus classification. It is load-bearing: both homotopy equivalences and the local contractibility of NU depend on identifying cut points with first conjugate points or non-unique maximizers, and on the assertion that a geodesic ceases to be maximizing beyond such a point. The reader and the stress-test note are right that the 'two distinct maximizing geodesics ⇒ no maximizing extension' direction is only cited, with omitted proofs in [1]. The reference to [18, Thm 2.9] (maximizers are pregeodesics) is not by itself enough; you need the standard corner/concatenation argument plus the reverse triangle inequality. That argument is almost certainly valid, but it is not written out, and Theorem 5.6 is too central to leave as a black box. This is a presentation gap, not a fatal flaw, but the author should provide a self-contained proof or a precise derivation in the paper.\n\nMinor note: a few technical lemmas come from the author's own preprints [16,17]; they look like standard tools, so I wouldn't hold that against the paper, but a referee should confirm they don't gate the main results.\n\nBottom line: the central derivation holds together, the topology arguments are coherent, and the reader's CONDITIONAL verdict is fair. I'd send this to a serious referee, and I'd cite it if I worked in Lorentzian weak KAM theory. Bring it to reading group.","headline":"First Lorentzian extension of Cannarsa-Cheng-Fathi's multiple-maximizer results, with solid new topology theorems; the main soft spot is the imported cut-locus classification in Theorem 5.6, which is a presentation gap rather than a fatal flaw.","tokens_in":737,"tokens_out":1139,"would_cite":true,"duration_ms":50970,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["53C50","37J51","53C22"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that in any globally hyperbolic spacetime the set of causally related pairs with multiple maximizing geodesics is locally contractible and has the same homotopy type as the causal future minus the Lorentzian Aubry set.","keywords":["Lorentzian geometry","globally hyperbolic spacetime","multiple maximizing geodesics","local contractibility","Lorentzian Aubry set","cut locus","Lax-Oleinik semigroup","Hamilton-Jacobi equations"],"falsifier":"Look for a globally hyperbolic spacetime with a future inextendible causal geodesic whose cut point is neither a first conjugate point nor the endpoint of a distinct maximizing geodesic; such a point would falsify Theorem 5.6 and with it the construction of the retraction from $J^+\\setminus\\mathcal A$ onto $\\mathrm{Cut}_M$. A more direct check would compute the homotopy type of $\\mathcal{NU}(M,g)$ and $J^+\\setminus\\mathcal A$ in a concrete cosmological model and find them different.","tokens_in":44322,"feed_emoji":"⏳","tokens_out":8549,"duration_ms":95616,"temperature":0.7,"pith_summary":"This paper extends a Riemannian theorem to the Lorentzian setting: in any globally hyperbolic spacetime, the set $\\mathcal{NU}(M,g)$ of pairs of causally related points joined by more than one maximizing geodesic is locally contractible. More strongly, it claims that $\\mathcal{NU}(M,g)$ has the same homotopy type as the causal future with a Lorentzian analogue of the Aubry set removed, and that the inclusions $\\mathcal{NU}(M,g)\\hookrightarrow \\mathrm{Cut}_M\\hookrightarrow J^+\\setminus\\mathcal A$ are homotopy equivalences. A reader should care because this ties the singularities of the Lorentzian distance, where maximizing geodesics branch, to the same objects that organize weak KAM theory and optimal transport on spacetimes, despite the failure of the usual Tonelli regularity assumptions. The paper's slice versions say that fixing one endpoint, the set of endpoints with multiple maximizers is locally contractible and homotopy equivalent to the future of that endpoint minus its Aubry set.","feed_headline":"Nonunique geodesics share the causal Aubry set's shape","feed_subtitle":"In globally hyperbolic spacetimes, pairs with several maximizing geodesics are topologically the same as the causal future minus the Aubry…","key_machinery":"The engine is the Lorentzian Lax-Oleinik semigroup built from the minimal action $c_t(x,y)=-t^{1/2}d(x,y)^{1/2}$, applied to the characteristic seed $\\chi_x$. The key regularity result is that $\\hat T_sT_{1+s}\\chi_x(y)$ is both locally semiconvex and locally semiconcave, hence $C^1$, on the chronological region, with a unique point $z$ attaining $\\hat T_sT_{1+s}\\chi_x(y)=T_{1+s}\\chi_x(z)-c_s(y,z)$. This uniqueness yields a continuous map $F(s,x,y)=z$ sending a chronological pair $(x,y)$ to a point $z$ that lies in the multiple-maximizer locus once $s>0$. Complementary to this, the classical cut-locus characterization (Theorem 5.6) and continuity of the cut-time $\\alpha$ are used to show that the cut locus is a strong deformation retract of $J^+\\setminus\\mathcal A$; combining the two gives the homotopy equivalences.","core_discovery":"The paper's central claim is that non-uniqueness of maximizing causal geodesics is not a local pathology but a global topological invariant: in any globally hyperbolic spacetime, the multiple-maximizer locus $\\mathcal{NU}(M,g)$ is locally contractible, and the inclusions $\\mathcal{NU}(M,g)\\hookrightarrow \\mathrm{Cut}_M$ and $\\mathrm{Cut}_M\\hookrightarrow J^+\\setminus\\mathcal A$ are homotopy equivalences, with the timelike analogue $\\mathcal{NU}_t(M,g)\\hookrightarrow I^+\\setminus\\mathcal A$. Here $\\mathrm{Cut}_M$ is the causal cut locus of the spacetime and $\\mathcal A$ is the Lorentzian Aubry set, the set of pairs $(x,y)$ lying on a future and past inextendible maximizing geodesic. The proof defines future Aubry sets $A(x)$ for single endpoints and proves the corresponding equivalences for each slice. Along the way it establishes that the composed Lax-Oleinik evolution $\\hat T_sT_{1+s}\\chi_x$ is $C^1$ on the chronological set and that its unique maximizer provides the continuous deformation pushing pairs onto the cut locus.","pith_inferences":["The homotopy-equivalence statement suggests that the non-uniqueness locus carries no extra topology beyond what is already visible in the Aubry set complement; a plausible test is to compute both sides in a concrete warped-product spacetime and compare Betti numbers.","The distinction between $\\mathcal A$ and $\\tilde{\\mathcal A}$ may matter for incomplete spacetimes: if causal geodesic completeness fails, the two definitions could yield different sets, and the theorem's dependence on the choice of line versus bi-infinite ray could be probed numerically in a spacetime with a null geodesic that stops being maximizing before it can be extended.","The $C^1$ regularity of $\\hat T_sT_{1+s}\\chi_x$ is proved only for the characteristic seed; if it extends to general lower semicontinuous initial data, the same deformation argument would give homotopy equivalences for singularities of more general Lorentzian Hamilton-Jacobi equations.","One could try to define a canonical 'cut-time' function $\\beta$ on $J^+\\setminus\\mathcal A$ as the infimum appearing in Lemma 5.25; making $\\beta$ explicit in integrable examples would yield a constructive retraction and possibly a new normal form for the causal future near the Aubry set."],"forward_implications":["If the claim holds, for every globally hyperbolic spacetime the topological invariants of $\\mathcal{NU}(M,g)$, such as path components, fundamental group, and homology, coincide with those of $J^+\\setminus\\mathcal A$.","For each fixed $x$, the slice of endpoints with multiple maximizing geodesics from $x$ is homotopy equivalent to $J^+(x)\\setminus A(x)$, so the singularity set of the future distance function has exactly the homotopy type of the complement of the future Aubry set.","Since the inclusions are homotopy equivalences, any continuous construction or invariant defined on the causal cut locus transfers to the non-uniqueness locus without loss; open covers and homotopies can be lifted back and forth between the two sets.","In causal geodesically complete spacetimes the two versions of the Lorentzian Aubry set agree, so the statement holds for both definitions and the proof simplifies.","The map $F$ gives an explicit homotopy proving local contractibility of $\\mathcal{NU}$, meaning small neighborhoods of any non-uniqueness pair can be contracted to one point while remaining inside the non-uniqueness set."],"supporting_citations":[{"why":"Supplies the Riemannian theorem this paper extends and the overall singularity-and-semigroup strategy.","marker":"[4]"},{"why":"Provides the cut-locus classification (Theorems 9.12, 9.15) and continuity of the cut-time function $\\alpha$ used in the retractions.","marker":"[1]"},{"why":"Gives the result that maximizing causal curves are pregeodesics (Theorem 2.9), closing gaps in the cut-locus argument.","marker":"[18]"},{"why":"Provides the Lax-Oleinik representation and semiconvexity arguments adapted to the characteristic seed.","marker":"[9]"},{"why":"Supplies the semiconcavity and semiconvexity toolbox used to control the composed semigroup.","marker":"[10]"},{"why":"Establishes the local semiconcavity of the Lorentzian cost on the chronological set.","marker":"[15]"},{"why":"Gives existence of maximizing geodesics in globally hyperbolic spacetimes used throughout the proofs.","marker":"[20]"},{"why":"Provides the result that unique super-differentiability implies differentiability for semiconcave functions, used to identify unique maximizers.","marker":"[21]"},{"why":"Supplies the uniform approximation by smooth semiconcave functions used in the semiconcavity proof.","marker":"[2]"},{"why":"Provides the noncompact-manifold approximation result for the semigroup argument.","marker":"[11]"}],"fun_headline_variants":["Multiple geodesics share the cut locus's homotopy type","Locus of multiple geodesics is locally contractible","Aubry set defines geodesic branching topology","Cut locus and future minus Aubry: same homotopy"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof imports, rather than proves, the classification that a causal geodesic stops being maximizing only at a first conjugate point or at a point where a second distinct maximizing geodesic arrives; if that classification had a counterexample in some globally hyperbolic spacetime, the deformation retractions that carry the main theorems would not be well defined.","fun_headline_variants_meta":{"raw":{"variants":["Multiple geodesics share the cut locus's homotopy type","Locus of multiple geodesics is locally contractible","Aubry set defines geodesic branching topology","Cut locus and future minus Aubry: same homotopy"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000957,"raw_usage":{"total_tokens":4064,"prompt_tokens":916,"completion_tokens":3148,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":532,"completion_tokens_details":{"reasoning_tokens":3082}},"tokens_in":532,"tokens_out":3148,"duration_ms":29098,"temperature":1.0,"reasoning_tokens":3082,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T11:20:47.997379+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Look for a globally hyperbolic spacetime with a future inextendible causal geodesic whose cut point is neither a first conjugate point nor the endpoint of a distinct maximizing geodesic; such a point would falsify Theorem 5.6 and with it the construction of the retraction from $J^+\\setminus\\mathcal A$ onto $\\mathrm{Cut}_M$. A more direct check would compute the homotopy type of $\\mathcal{NU}(M,g)$ and $J^+\\setminus\\mathcal A$ in a concrete cosmological model and find them different.","supporting_citations":[{"cited_title":"Singularities of solu- tions of time dependent Hamilton-Jacobi equations","cited_arxiv_id":null,"evidence_quote":"Supplies the Riemannian theorem this paper extends and the overall singularity-and-semigroup strategy."},{"cited_title":"Global Lorentzian Geometry, Second Edition","cited_arxiv_id":null,"evidence_quote":"Provides the cut-locus classification (Theorems 9.12, 9.15) and continuity of the cut-time function $\\alpha$ used in the retractions."},{"cited_title":"Lorentzian causality theory","cited_arxiv_id":null,"evidence_quote":"Gives the result that maximizing causal curves are pregeodesics (Theorem 2.9), closing gaps in the cut-locus argument."},{"cited_title":"Viscosity solutions of the Hamilton-Jacobi equation on a noncompact manifold","cited_arxiv_id":null,"evidence_quote":"Provides the Lax-Oleinik representation and semiconvexity arguments adapted to the characteristic seed."},{"cited_title":"Optimal transportation on non-compact manifolds","cited_arxiv_id":null,"evidence_quote":"Supplies the semiconcavity and semiconvexity toolbox used to control the composed semigroup."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the local semiconcavity of the Lorentzian cost on the chronological set."},{"cited_title":"Semi-Riemannian geometry with applications to relativity: Volume 103","cited_arxiv_id":null,"evidence_quote":"Gives existence of maximizing geodesics in globally hyperbolic spacetimes used throughout the proofs."},{"cited_title":"Optimal Transport: Old and New","cited_arxiv_id":null,"evidence_quote":"Provides the result that unique super-differentiability implies differentiability for semiconcave functions, used to identify unique maximizers."},{"cited_title":"Existence of C 1,1 critical sub-solutions of the Hamilton- Jacobi equation on compact manifolds","cited_arxiv_id":null,"evidence_quote":"Supplies the uniform approximation by smooth semiconcave functions used in the semiconcavity proof."},{"cited_title":"On the Hausdorff di- mension of the Mather quotient","cited_arxiv_id":null,"evidence_quote":"Provides the noncompact-manifold approximation result for the semigroup argument."}],"review_version":1}