{"id":"f5af4034-6545-4e3f-86d7-420f97b8af42","arxiv_id":"2508.20062","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Tropical linear series on metric graphs are locally Bergman fans of matroids, yielding an exact condition for canonical tropicalizations to fill the realizable locus.","lead":"Tropical linear series on metric graphs are shown to be locally identical to Bergman fans of matroids at their generic divisors. The paper also gives an exact condition under which one curve's canonical tropical linear series fills the entire realizable canonical divisor locus.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1.14's local-coordinate injectivity rests on an unproved slope constraint; if it fails, Star(D) need not be the Bergman fan.","rationale":"The reader's weakest assumption identifies the same step: the local-coordinate injectivity in Theorem 1.14. My reading of the proof confirms that this is the most load-bearing point: everything after the construction of the flat lattice M_Σ depends on the evaluation map being an embedding onto a neighborhood of 0 in the Bergman fan. The proof as written compresses the justification of the slope constraint into one sentence, and the constraint is genuinely nontrivial because support-count local maximality is a global property of the divisor D + div(φ), not a direct pointwise condition on slope vectors. I do not claim the theorem is false; the statement is plausible and consistent with known results, but the proof as written has a gap at exactly this point. The omitted case analysis in Example 8.4 flagged by the reader is real but peripheral; this slope-constraint issue is more central. A concrete computational check on a small matroidal example would either confirm the assertion and close the gap, or produce a counterexample and invalidate the theorem. Since the reader's verdict is already CONDITIONAL, my concern does not move the verdict, but it does sharpen the condition: the revision should supply a complete proof of the slope-constraint lemma or weaken Theorem 1.14 accordingly.","tokens_in":41202,"tokens_out":12048,"duration_ms":155672,"concrete_test":"Choose a concrete nondegenerate divisor in a nontrivial tropical linear series, e.g. the matroidal series Φ(Trop(U_{2,4})) on an interval with D of degree 2 from Example 6.4, or a degree-2 divisor on a 3-valent graph. Enumerate all φ = min_i(c_i + φ_i) with c_i in a small box, compute D + div(φ), and verify directly: (a) at every x ∈ supp(D), the nonzero outgoing slopes of φ are all equal to D(x) and occur in at most one tangent direction; (b) the evaluation vector (φ(p_1),...,φ(p_s)) at one point in each component of Γ∖supp(D) determines φ uniquely on the box. If any φ in the box violates (a) or (b), the injectivity step in Theorem 1.14 fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central step in the proof of Theorem 1.14 is the claim that, for a nondegenerate D and a sufficiently small φ in Σ with min φ = 0, the local maximality of #supp forces, at every x ∈ supp(D), each nonzero outgoing slope of φ along a tangent direction to equal D(x), and at most one such tangent direction to carry nonzero slope. This is asserted without proof in the middle of the proof of Theorem 1.14, and it is exactly what makes the evaluation map injective and gives the identification Star(D) ≅ B(M_Σ). The implication is not immediate: when D(x) > 1, slopes can cancel at x (e.g. two tangent directions with slopes +a and −a give ord_x(φ) = 0), and the condition that D + div(φ) does not increase #supp constrains the divisor as a configuration of points, not just the slope vector at x. A proof would need to rule out such cancellations using the effective divisor condition and the local-minimality of the valence-1 degree, accounting for coincident support points. Without this, the map from |Σ| near D to R^E/(1,...,1) may fail to be injective, and the local cone structure could contain geometry beyond the flats {F_φ}. This is the load-bearing bridge from the combinatorial lattice of minimizers to the actual Bergman fan structure.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a theory of tropical linear series on metric graphs, defined as finitely generated tropical submodules Σ ⊆ R(D) satisfying rind(Σ) = rBN(Σ) + 1. The main structural result is Theorem 1.14: at a nondegenerate divisor D, the local geometry of |Σ| is the Bergman fan of a matroid MΣ, whose flats are read off from the minimizers of functions in Σ. The paper also proves pure dimensionality (Theorem 1.5), relates matroidal linear series to valuated matroids (Theorem 1.7, Theorem 6.1), characterizes when the tropicalization of the canonical linear series of a single curve equals the realizable locus Real(|KΓ|) (Theorem 1.15), proves that every loopless matroid arises as a local matroid (Theorem 1.18), and revisits Cartwright divisors and matroid adjoints (Theorems 7.1, 7.3). The exposition includes many examples, a streamlined proof of Dupraz's equidimensionality theorem, and an appendix proving that tropicalizations of algebraic linear series are matroidal.","tokens_in":41467,"tokens_out":2911,"duration_ms":35839,"significance":"If the proofs are completed, this is a significant contribution: it gives a unified local-global picture in which tropical linear series are controlled by matroids and valuated matroids, and it resolves a previous question about equidimensionality. The paper is largely self-contained: Appendix A gives a full proof that tropicalizations are matroidal; Corollary 4.10 is an elegant maximality statement; and the explicit constructions of non-realizable matroidal linear series (Examples 6.6 and 6.8) are valuable. However, the central local structure theorem (Theorem 1.14) contains an unproved slope-injectivity assertion that is load-bearing, and several auxiliary claims are terse. These gaps are patchable in principle, but they need to be addressed before the main theorem is established.","major_comments":[{"comment":"The injectivity step, beginning 'Since D has locally maximal support, the slope of φ on any tangent vector ζ at x is either 0 or the multiplicity D(x)', is asserted without proof. This is the bridge from the lattice of minimizers to the actual local fan structure: it is exactly what makes the evaluation map injective and identifies Star(D) with B(MΣ). The implication is not immediate when D(x) > 1, because two tangent directions with slopes +a and −a contribute ord_x(φ) = 0 and can be hidden from the divisor condition. The proof must rule out such cancellations using effectiveness of D + div(φ) and the local minimality of the valence-1 degree, including coincident support points. Without a detailed argument, Theorem 1.14—and hence the paper's central claim—is not fully established.","section":"§4, proof of Theorem 1.14"},{"comment":"The complement-closure step in the proof of Lemma 3.8 is very terse. Starting with φ ∈ Σ, the paper considers D_ε = div(min{φ, ε}) and asserts that, because D lies in the relative interior of a maximal face τ, 'the opposite of any such chip-firing move is well-defined in Σ'. This is a nontrivial statement: it requires that for every sufficiently small such ε there exists φ' ∈ Σ with div(φ') equal to the opposite divisor, and that the resulting construction is compatible with the face τ. This step is needed to conclude that L ∪ {E} is closed under complements and hence is a Boolean lattice, which in turn is used to bound dim |Σ|. Please expand this argument with a precise local-coordinate or chip-firing proof.","section":"§3, Lemma 3.8"},{"comment":"The proof of polyhedrality of Real(|KΓ|) invokes elimination of quantifiers and definability in algebraically closed valued fields, then compactness of S_Γ to obtain a closed polyhedral set. The model-theoretic argument is sketched in four sentences. Since this proposition is a key input to Theorem 1.15, either give a more detailed derivation of the definable/polyhedral statement or cite a specific reference for the quantifier-elimination step in this exact setting. As written, the reader cannot check the claim that the image is a finite Boolean combination of polyhedra rather than a more general definable set.","section":"§5, Proposition 5.2"}],"minor_comments":[{"comment":"The text explicitly says 'We omit the cumbersome case analysis' when asserting that every rank-1 degree-3 divisor on the loop of loops is equivalent to a divisor of the form v1 + w3 + w. For a self-contained example, this case analysis should be supplied or at least summarized, especially since the example is used to illustrate uniqueness of a tropical linear series.","section":"§8, Example 8.4"},{"comment":"In the derivation of equation (1), the dimension inequality rind(Σ) = dim |Σ| + 1 is used before Corollary 3.9 is stated. This is acceptable logically since the text refers forward, but the forward reference should be explicit at that point.","section":"§1, Remark 1.3"},{"comment":"In the proof of uniqueness and continuity of the section σ, the existence of an open dense subset U' over which π is a homeomorphism is asserted without proof. This follows from piecewise-linearity and equality of dimensions, but a brief justification would improve readability.","section":"§6, Theorem 6.1"},{"comment":"The notation eΦ is used for the evaluation map in the proof of Theorem 1.14, but it is never defined in the text; presumably it should be the map φ ↦ (φ(p1), …, φ(p_s)). Please clarify.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The central issue is the unproved slope constraint in Theorem 1.14. If the authors can supply a rigorous proof of that injectivity claim—and patch the terse complement-closure step in Lemma 3.8—the paper would be in good shape. The Theorem 1.14 gap is not a matter of disagreement with consensus; it is a missing technical verification in the manuscript itself. I would not recommend rejection outright, because the surrounding framework and the statement are plausible and the rest of the paper is coherent. But I would not accept before seeing a complete proof."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my take. The paper gives a clean definition of tropical linear series (r_ind = r_BN + 1) and then proves that near a nondegenerate divisor, the star of the projectivization is the Bergman fan of a matroid — the local matroid. This is genuinely new and, if correct, is the right local structure theorem for the subject. It also shows every loopless matroid occurs as a local matroid, characterizes when the tropicalized canonical series equals the realizable locus, and gives new non-realizability obstructions via local matroids. The writing is honest: it explicitly says pure dimensionality was already in Dupraz's thesis and [AGG25], and it supersedes [JP22]. That is the right way to handle prior work.\n\nThe soft spot is in the proof of Theorem 1.14. To show the evaluation map is injective and lands on the Bergman fan, the proof asserts that for small φ with min φ = 0, at each x in supp(D), every nonzero outgoing slope of φ equals D(x), and at most one tangent direction carries nonzero slope. This is stated without proof. It is not obviously implied by local maximality of #supp and local minimality of valence-1 degree: when D(x) > 1, slopes can cancel, and the effective divisor condition constrains the whole configuration, not just the slope vector at x. This is exactly the step that identifies Star(D) with B(M). If it fails, the local matroid theorem is not established. I don't think the claim is false, but the proof needs a real argument here.\n\nThere are smaller issues: Lemma 3.8's complement-closure step is compressed, and Example 8.4 explicitly omits a case analysis, which is not acceptable for a claimed classification in an example. Proposition 5.2 invokes elimination of quantifiers; that's legitimate, but some readers will want a combinatorial proof.\n\nI'd send this to a serious referee. The paper is important enough, and the gap is localized. The referee should be told to check the slope constraint carefully. With that fixed, this would be a strong paper. I would bring it to the reading group now, because the gap is pedagogically useful and the examples are thought-provoking.","headline":"A substantial and honest paper on tropical linear series with a genuinely new local matroid theorem, but the proof of Theorem 1.14 has an unproved slope constraint that needs attention before I'd trust the central identification.","tokens_in":42032,"tokens_out":7282,"would_cite":true,"duration_ms":86099,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14T10","14T15","05B35","14H51"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that every tropical linear series is locally the Bergman fan of a matroid, and uses that to characterize when canonical linear series tropicalize to the realizable locus.","keywords":["tropical linear series","metric graphs","Baker–Norine rank","tropical independence","valuated matroids","Bergman fans","tropical canonical divisors","matroid realizability"],"falsifier":"Take a tropical linear series on an interval such as the degree-2 complete series R(2v) with the uniform matroid U2,3 parametrization, and examine a divisor in the interior of |Σ|. If the link of that divisor is not the three-ray Bergman fan of U2,3, or if two distinct nearby divisors give the same evaluation at points in each component outside the support, then Theorem 1.14 fails.","tokens_in":41055,"feed_emoji":"📐","tokens_out":5859,"duration_ms":59744,"temperature":0.7,"pith_summary":"The paper defines a tropical linear series as a finitely generated tropical module of piecewise-linear functions on a metric graph whose tropical-independence rank is exactly one more than its Baker–Norine rank. Its central result is that at any nondegenerate divisor in the series, the local shape of the series is completely controlled by a matroid: the flats are read off from where functions attain their minima, and the local fan is the Bergman fan of that matroid. This gives a precise way to test realizability of linear series, since local matroids of realizable series must be realizable, and it yields a complete criterion for when the tropicalization of the canonical linear series on a single curve fills the locus of realizable canonical divisors. The paper also shows that every loopless matroid occurs as such a local matroid, and it uses matroidal examples to show that the wider class of tropical linear series can fail the recursive incidence properties that algebraic linear series satisfy.","feed_headline":"Local shape of every tropical linear series is a matroid's Bergman fan","feed_subtitle":"It pins the two tropical ranks to matroids and settles when canonical series fill the realizable locus.","key_machinery":"The central object is the local matroid M_Σ. For a tropical linear series Σ ⊆ R(D) and a nondegenerate divisor D, the ground set is the set E of connected components of Γ minus the support of D; each function φ ∈ Σ contributes the flat F_φ, namely the components not contained in φ's minimizer. The key mechanism is that these sets form a matroid lattice of flats, and that evaluation at one point per component gives local coordinates identifying Star(D) with the Bergman fan of M_Σ. The technical condition of big minimizers—every minimizer contains a whole component—allows the matroid to be defined even when D is not in |Σ|. Valuated matroids enter as the tropical modules that parametrize matro","core_discovery":"The paper's main theorem states that if Σ is a tropical linear series of dimension r and D is a nondegenerate divisor in |Σ|, then the collection {F_φ : φ ∈ Σ} ∪ {E}, where E is the set of connected components of Γ ∖ supp(D) and F_φ is the set of components not contained in the minimizer of φ, is the lattice of flats of a matroid M_Σ of rank r+1. Moreover, evaluation at one point in each component embeds the local space Star(D) into R^E/(1,…,1), and the image is exactly the support of the Bergman fan of M_Σ. The same construction gives an invariant local matroid whenever the series has big minimizers, i.e., every minimizer contains a whole connected component. The paper further proves that e","pith_inferences":["If the local-matroid construction extends to degenerate divisors, the local structure of a tropical linear series would become a stratification by matroid quotients, potentially leading to a global parametrization and answering Question 9.1 for intervals and loops.","Because every loopless matroid is a local matroid on an interval, tropical linear series on an interval could serve as a universal combinatorial model in which any Bergman fan embeds as a local fan inside a complete linear system.","Theorem 1.15 suggests a computational recipe: compute the dimension of Real(|KΓ|) from the Möller–Ulirsch–Werner conditions; equality with g−1 then identifies exactly the curves whose canonical tropicalization is the full realizable locus.","The Vámos-based examples indicate that failure of the recursive incidence properties is a matroidal phenomenon rather than a special pathology of tropical linear series, and such examples could be used to test whether every tropical linear series is matroidal."],"forward_implications":["Every tropical linear series has pure dimension equal to its Baker–Norine rank, so the projectivized series has no higher-dimensional whiskers.","If a tropical linear series is realizable, then every local matroid at a nondegenerate divisor is realizable; a non-realizable local matroid is therefore an explicit obstruction to lifting to an algebraic linear series.","Every loopless matroid appears as the local matroid of a tropical linear series at a nondegenerate divisor, both on an interval and on a loop, so all Bergman fans occur as local fans in divisor spaces.","For a curve X with skeleton Γ in equicharacteristic zero, Trop(|K_X|) = Real(|KΓ|) if and only if Real(|KΓ|) has dimension g−1; when this holds, every curve with that skeleton has the same canonical tropicalization.","Inclusions among tropical linear series are rigid: if Σ ⊆ Σ′ and both are tropical linear series of the same dimension, then Σ = Σ′, giving a maximality property for these modules."],"supporting_citations":[{"why":"Supplies the Baker–Norine rank and the graph Riemann–Roch theorem, the rank notion on which the definition of tropical linear series is built.","marker":"[BN07]"},{"why":"Establishes finite generation of R(D) and the polyhedral structure of complete linear systems, and gives the cographic-matroid observation for canonical divisors.","marker":"[HMY12]"},{"why":"Provides the certificate criterion for tropical independence and the proof that tropicalizations of linear series are finitely generated tropical modules, underlying Theorem 1.7.","marker":"[FJP25]"},{"why":"Contains the original proof of pure dimensionality of tropical linear series, streamlined here as Theorem 1.5, and the polyhedrality of the realizable canonical-divisor locus.","marker":"[Dup24]"},{"why":"Determines the realizable locus Real(|KΓ|) of tropical canonical divisors, which is the object characterized in Theorem 1.15.","marker":"[MUW21]"},{"why":"Introduces Cartwright divisors on Levi graphs of rank-3 matroids and proves their realizability criterion, which the paper extends using local matroids and adjoints.","marker":"[Car15]"},{"why":"Supplies the equivalence between tropical rank and dimension, and the theory of initial matroids and Bergman fans used in the local-structure arguments.","marker":"[MS15]"},{"why":"Provides Vámos-matroid examples where tropical linear spaces fail incidence properties, used to construct non-recursive tropical linear series.","marker":"[Wan24]"}],"fun_headline_variants":["Tropical linear series: local fans are matroid Bergman fans","Every tropical linear series has a local matroid fan","Bergman fans appear locally in tropical linear series","Matroid fans govern tropical linear series locally"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The proof assumes that, near a nondegenerate divisor, each function in the tropical linear series is completely determined by its values at one chosen point in each connected component of the graph outside the divisor's support; the two nondegeneracy conditions—maximal support count and minimal valence-1 degree—are what make this local-coordinate injectivity hold.","fun_headline_variants_meta":{"raw":{"variants":["Tropical linear series: local fans are matroid Bergman fans","Every tropical linear series has a local matroid fan","Bergman fans appear locally in tropical linear series","Matroid fans govern tropical linear series locally"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000201,"raw_usage":{"total_tokens":1188,"prompt_tokens":692,"completion_tokens":496,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":436,"completion_tokens_details":{"reasoning_tokens":433}},"tokens_in":436,"tokens_out":496,"duration_ms":6184,"temperature":1.0,"reasoning_tokens":433,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T15:14:28.361697+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a tropical linear series on an interval such as the degree-2 complete series R(2v) with the uniform matroid U2,3 parametrization, and examine a divisor in the interior of |Σ|. If the link of that divisor is not the three-ray Bergman fan of U2,3, or if two distinct nearby divisors give the same evaluation at points in each component outside the support, then Theorem 1.14 fails.","supporting_citations":[],"review_version":1}