{"id":"15d91584-e0b8-462e-aa5f-c1427e04918a","arxiv_id":"2412.18206","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Koszulity of an indiscretely based category algebra is equivalent to the local bouquet property of its factorization spaces.","lead":"Category algebras carry a strong homological regularity condition called Koszulity; this paper proves that for a broad class of them, Koszulity is exactly equivalent to a topological property of spaces built from the categories' morphisms. The result creates a new algebra-topology bridge and, applied to toric geometry, tells when dual collections of line bundles can be shifted to become strong.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proposition 2.40 is false as stated for length-zero morphisms: length-0 isomorphisms contribute \\tilde H^{-1}(∅)=k to Ext^1(A_0,A_0)_0, which is actually 0 (e.g., k[x]).","rationale":"The reader identified the 'indiscretely based' assumption as the weakest point, but my concern is different and more specific: Proposition 2.40, the load-bearing computational result, is literally false as stated because it includes length-zero isomorphisms in the direct sum. For p an isomorphism, BC(p) is empty, and the paper's own convention \\tilde H^{-1}(∅)=k produces a spurious contribution to Ext^1(A_0,A_0)_0. A concrete counterexample is the one-object category k[x]: Proposition 2.40 would predict Ext^1(k,k)_0 = k, while the actual value is 0. The same issue appears for i=0, where the formula would predict Ext^0(S_v,S_v)=0 instead of k. This does not necessarily destroy the main theorem, because the offending terms appear only when i=0 or i=1 and n=0, and Koszulity at n=0 is automatic; however, the proof of Theorem 2.66 as written relies on the full equality and is therefore invalid. The theorem and surrounding results are likely repairable by restricting the sum to non-isomorphisms and handling low-degree Ext groups separately, but this is a genuine correctness issue that a reader should not have to infer. Hence the verdict should be conditional acceptance rather than unconditional acceptance.","tokens_in":37687,"tokens_out":27252,"duration_ms":253564,"concrete_test":"Compute both sides of Proposition 2.40 for the one-object category with kC = k[x] and p = id_v: the LHS Ext^1_{k[x]}(k,k)_0 is 0, while the RHS sums over p with l(p)=0 giving \\tilde H^{-1}(BC(id_v)) = \\tilde H^{-1}(∅) = k, a contradiction. Also check i=0, n=0: the LHS Hom_{k[x]}(k,k) = k, while the RHS \\tilde H^{-2}(∅) = 0. If the paper is revised to sum only over non-isomorphism morphisms (l(p)>0) and to add explicit augmentation terms for i=0,1, rerun these two checks to confirm they become equalities.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Proposition 2.40 claims Ext^i_{kC}(S_w,S_v)_{-n} = \\bigoplus_{p:l(p)=n,t(p)=v,h(p)=w} \\tilde H^{i-2}(BC(p)). Take C with one object v and morphisms {x^m}, so kC ≅ k[x] (Example 2.6). For i=1, n=0, the only p with l(p)=0 is id_v. There is no nontrivial factorization of id_v, so BC(id_v)=∅ by Definition 2.31. With the convention \\tilde H^{-1}(∅)=k used in Proposition 2.62, the RHS is k. But Ext^1_{k[x]}(k,k)_0 = 0, as seen from the standard resolution 0 → k[x](-1) → k[x] → k → 0. The same failure occurs for i=0, n=0: the LHS Hom(S_v,S_v) = k, while the RHS is \\tilde H^{-2}(∅)=0. The direct sum in Proposition 2.40 must range over non-isomorphism morphisms (equivalently, over cells of the reduced nerve), and the augmentation terms for i=0 and i=1 must be handled separately. As stated, the proposition is false, and since Theorem 2.66 invokes this equality for all i,n, the proof of the main theorem is invalid as written. The theorem may be salvageable by restricting the sum to l(p)>0 and treating i=0,1 explicitly, but the paper requires correction.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies category algebras of N-graded categories that are \"indiscretely based\" (no nontrivial automorphisms; the degree-zero connected components are indiscrete categories). It introduces a reduced nerve BC and factorization spaces BC(p) and claims a formula (Proposition 2.40) expressing Ext^i between simple functors as a direct sum of reduced cohomology groups \\tilde H^{i-2}(BC(p)). From this it derives the main theorem (Theorem 2.66): kC is Koszul if and only if C is locally bouquet. It then defines almost discrete fibrations, proves that Reiner–Stamate equivalence relations on posets are exactly such fibrations, recovers the Reiner–Stamate Koszulity criterion, and applies the framework to homotopy path algebras, Bondal–Thomsen algebras, and full strong exceptional collections of line bundles on toric varieties.","tokens_in":37935,"tokens_out":16018,"duration_ms":156421,"significance":"The conceptual package is valuable: reindexing the normalized standard resolution by cells of a reduced nerve is a natural idea, and the paper demonstrates its use by recovering the poset case, the Reiner–Stamate reduced incidence algebra theorem, and the Favero–Huang sufficient condition, while also giving new toric consequences. If the main theorem is correct, it provides a clean topological criterion for Koszulity in a broad class of algebras. However, the central computational proposition has a low-degree defect, and the main theorem is therefore not proved as written. The paper is worth pursuing after a focused revision that corrects the statement and proof of Proposition 2.40 and the statements that depend on it.","major_comments":[{"comment":"The displayed equality is false as stated for length-zero morphisms and for i=0,1. Take C with one object v and morphisms {x^n:n≥0}, so kC≅k[x] (Example 2.6). For p=id_v, BC(p)=∅ because there are no nontrivial factorizations of an isomorphism. With the convention \\tilde H^{-1}(∅)=k adopted in the preamble to Proposition 2.62, the right-hand side for (i,n)=(1,0) contains the summand k from p=id_v, but Ext^1_{k[x]}(k,k)_0=0, as read off from the standard resolution 0→k[x](-1)→k[x]→k→0. For (i,n)=(0,0) the right-hand side is \\tilde H^{-2}(∅)=0, while the left-hand side is Hom_{kC}(S_v,S_v)=k. Thus Proposition 2.40 cannot be used for all i,n as Theorem 2.66, Corollary 2.44, and Proposition 2.62 do. The proof appears to require restricting the direct sum to non-isomorphism morphisms and treating the degree-zero Ext^0 and Ext^1 terms separately; as written, the proposition is false.","section":"Section 2.3, Proposition 2.40"},{"comment":"Condition (3), that a morphism is indecomposable if and only if it has length 1, is false for length-zero morphisms: every isomorphism, in particular every identity, is indecomposable because a nontrivial factorization cannot compose to an isomorphism. Moreover condition (2) cannot be interpreted as including p=id, since then \\tilde H^{-1}(BC(id))=\\tilde H^{-1}(∅)=k for every object, so (2) would fail for every category and would force (1) to fail, contradicting standard examples such as k[x]. The proposition and its proof need to be limited to positive-length morphisms and to exclude the identity from the vanishing condition. Proposition 2.64 inherits the same problem in its hypothesis that BC(p) is non-empty for all p with l(p)≠1.","section":"Section 2.4.4, Proposition 2.62"},{"comment":"The step that a direct sum of \\tilde H^{i-2}(BC(p)) over all p of length n vanishes if and only if each BC(p) is bouquet is compressed and needs an explicit dimension argument. For a morphism p with l(p)=n, any nontrivial factorization has each factor a non-isomorphism, hence positive length, so a cell of dimension r in BC(p) corresponds to r+2 factors and therefore r+2≤n; thus dim BC(p)≤n-2. Reduced cohomology in degrees above the dimension vanishes automatically, so the vanishing required for Koszulity, namely \\tilde H^{i-2}(BC(p))=0 for i≠n, is equivalent to \\tilde H^j(BC(p))=0 for j<dim BC(p) in the relevant range. This should be stated explicitly; the current one-line equivalence hides the point and, together with the Proposition 2.40 issue, makes the proof of Theorem 2.66 incomplete.","section":"Section 2.4.4, proof of Theorem 2.66"},{"comment":"Proposition 2.16 is stated without the indiscretely based hypothesis and is false in general. If C is a one-object category whose morphisms form a nontrivial finite group G, then S_v is the trivial k[G]-module, which is not simple when the characteristic of k divides |G|, even though every morphism of C is an isomorphism. The proof implicitly uses the uniqueness of isomorphisms that is supplied later by the indiscretely based hypothesis. The proposition should be stated under that hypothesis or made conditional on it.","section":"Section 2.1, Proposition 2.16"}],"minor_comments":[{"comment":"The symbol Q0 is used without definition; it should be Ob(C).","section":"Section 2.4.4, proof of Proposition 2.62"},{"comment":"The notation \\tilde H^i(Rn(D,E)) is missing the subscript on R^n; it should refer to R^n_{(D,E)} as defined in Section 4.2.3.","section":"Theorem 1.4 and Theorem 4.53"},{"comment":"The phrase \"they are in discretely based\" appears to be a typo for \"indiscretely based\".","section":"Example 2.22"},{"comment":"The name Reiner is misspelled as \"Riener\" in the opening sentence.","section":"Section 3, opening paragraph"}],"recommendation":"major_revision","confidential_remarks":"The skeptic's counterexample is decisive: Proposition 2.40 is false as stated, and the main theorem depends on it. The flaw is localized to identity/length-zero contributions and to the low-degree Ext groups, and it appears repairable by a careful restatement that separates isomorphisms from non-isomorphisms and handles Ext^0 and Ext^1 in degree zero separately. I therefore recommend major revision rather than rejection. The reader's accept verdict seems too optimistic in light of this load-bearing error. The authors should also restate Proposition 2.62 for positive-length morphisms and make Proposition 2.16 conditional on the indiscretely based hypothesis."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth a look, but the main computational proposition has a bug that needs fixing before the main theorem can be trusted. Proposition 2.40 claims a direct sum over all p with l(p)=n of reduced cohomology of BC(p). For n=0, p is an isomorphism, and BC(p) is empty. For i=1 that gives \\tilde H^{-1}(∅)=k, but Ext^1(k[x])(k,k)_0=0. The same issue hits i=0,n=0, where the RHS is 0 but Hom is k. The proof via Lemma 2.39 only covers non-isomorphism p and cells of dimension at least 2, so the statement overreaches. This is not cosmetic: Theorem 2.66 invokes the equality for all i,n, and Proposition 2.62 inherits the problem. The fix is straightforward: restrict the sum to p of positive length (or non-isomorphisms) and handle Ext^0 and Ext^1 separately. I expect the theorem survives, but the paper as written is not correct.\n\nWhat is genuinely new and good: the idea of using the reduced nerve and factorization spaces to get a topological handle on Ext between simples is a real step beyond the poset cases; the locally bouquet criterion is a natural common generalization of Reisner-style conditions for incidence algebras and the sufficient condition in FH22; and the almost discrete fibration transfer (Corollary 2.78) gives a clean functorial mechanism, which then produces the Reiner–Stamate correspondence in Section 3. The applications to Bondal–Thomsen HPAs and toric line bundle collections are also useful, and the paper is honest about relying on prior work.\n\nOther soft spots: Proposition 2.62(3) states an iff that is only an implication (a length-one morphism is indecomposable, but the converse needs the previous condition), and the proof of Proposition 4.49 is compressed—it leans on star-shapedness and a homeomorphism claim that is plausible but not fully shown. These are minor compared to the Prop 2.40 issue.\n\nWho is this for? People working on Koszul algebras, incidence algebras, and homotopy path algebras will want to read it, but they should treat the current version as a draft. It deserves a serious referee, not a desk reject, because the framework and applications are valuable and the flaw is local and fixable. I would not cite it in its present form; once corrected, likely yes.","headline":"Strong framework, but Proposition 2.40 is false for length-zero morphisms, and the main theorem's proof needs repair before the paper is correct as stated.","tokens_in":38537,"tokens_out":5718,"would_cite":false,"duration_ms":55044,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["16S37","16E30","18G35","06A11"],"pacs":[],"model":"deepseek-v4-flash","headline":"The Koszulity of a category algebra is equivalent to a topological bouquet condition on its factorization spaces.","keywords":["Koszul algebras","category algebras","locally bouquet","factorization spaces","reduced nerve","incidence algebras","homotopy path algebras","exceptional collections"],"falsifier":"Take the entrance path category of the two-dimensional torus with the four-strata cube stratification from Example 4.29/4.40; the interval (~v,~w) is two disjoint arcs, so BC(p) for the corresponding path is disconnected and \\tilde $H^{0}$(BC(p)) \\neq 0. The theorem predicts $Ext^{2}$(S_w,S_v)_{-n} \\neq 0 for some n \\neq 2, hence non-Koszulity; a direct computation of $Ext^{2}$ from the quiver with the single cubic relation would settle whether the prediction holds.","tokens_in":37401,"feed_emoji":"🧩","tokens_out":4897,"duration_ms":49264,"temperature":0.7,"pith_summary":"This paper proves that for a large class of graded categories, the Koszul property of the category algebra is exactly a topological property of the category. The topological invariant is a family of spaces BC(p), one for each morphism p, built from the reduced nerve of the nontrivial factorizations of p. The main theorem says kC is Koszul if and only if each BC(p) is a bouquet, meaning it has no reduced cohomology below its dimension. This unifies and recovers the classical theorem that incidence algebras of graded posets are Koszul exactly when the poset is locally Cohen-Macaulay, and it extends to homotopy path algebras and to endomorphism algebras of line bundles on toric varieties.","feed_headline":"Koszulity is a topological condition on category algebras","feed_subtitle":"Ext groups reduce to cohomology of factorization spaces, giving a complete criterion for Koszulity.","key_machinery":"The load-bearing object is the reduced nerve \\bar N(C) of an indiscretely based category and, for each morphism p, the factorization space BC(p): the geometric realization of the semi-simplicial set of nontrivial factorizations p = f_0 \\circ \\dots \\circ f_{n+1} modulo the relation that inserts isomorphisms between adjacent factors. This space carries a cellular projective resolution of the diagonal bimodule, and the complex computing Ext is the CW cohomology chain complex of BC(p). The identity doing the work is the formula expressing Ext^i(S_w,S_v)_{-n} as a direct sum of reduced cohomology groups \\tilde $H^{{i-2}}$(BC(p)) over all paths of length n from v to w.","core_discovery":"The central discovery is a topological formula for Ext groups between simple representations of an indiscretely based category: Ext^i_{kC}(S_w,S_v)_{-n} = \\bigoplus_{p: l(p)=n, t(p)=v, h(p)=w} \\tilde $H^{{i-2}}$(BC(p)), where BC(p) is the geometric realization of the factorization space of p. Consequently Koszulity, which by the standard criterion is the vanishing of Ext^i(A_0,A_0)_{-n} for i \\neq n, becomes the statement that every BC(p) is cohomologically bouquet. The argument runs through a cellular projective resolution of the diagonal bimodule whose k-cells are indexed by cells of the reduced nerve, obtained from the normalized standard resolution after passing to a skeletal category. This reduces a purely algebraic property to a collection of topological vanishing statements.","pith_inferences":["The Ext formula suggests that Koszulity could be tested algorithmically by computing reduced cohomology of finite semi-simplicial sets, which is more tractable than resolving the algebra directly.","The paper explicitly notes that nontrivial automorphisms would bring representation theory of automorphism groups into the story; a plausible extension would replace the simple functors by simples with equivariant structure and formulate a twisted, character-dependent bouquet condition.","Because Koszulity passes to saturated subcategories, the theorem gives a quick obstruction: any non-locally-bouquet subcategory inside a Koszul category would force non-Koszulity, a shortcut for identifying examples like the Hirzebruch surface F1.","The toric application indicates that strongness of a shifted dual exceptional collection can be certified by local Cohen-Macaulayness of monomial posets, which may be easier to check than computing all derived Hom spaces."],"forward_implications":["For every graded poset P, the incidence algebra kP is Koszul if and only if P is locally Cohen-Macaulay, recovering the known theorem as a corollary.","Almost discrete fibrations preserve Koszulity; Reiner-Stamate equivalence relations on graded posets are exactly such fibrations, so Koszulity of an incidence algebra and its reduced incidence algebra coincide.","A graded homotopy path algebra is Koszul if and only if its path poset is locally Cohen-Macaulay, turning a homological property into a purely combinatorial one.","For Bondal-Thomsen homotopy path algebras, Koszulity is equivalent to the vanishing of reduced cohomology of certain unions of strata associated to open intervals in the induced stratification.","For a full strong exceptional collection of line bundles on a toric variety with a graded ordering, Koszulity is equivalent to the existence of shifts making the shifted dual exceptional collection strong."],"supporting_citations":[{"why":"Supplies the normalized standard resolution of the diagonal bimodule and the equivalence between category representations and modules used to set up the Ext computation.","marker":"[Mit72]"},{"why":"Provides the category-algebra module equivalence and the projective representable functors that underlie the resolution.","marker":"[Web07]"},{"why":"Supplies the Koszulity criterion Ext^i(A_0,A_0)_{-n}=0 for i\\neq n and the quadratic-generation facts used in the topological interpretation.","marker":"[BGS96]"},{"why":"Gives the bouquet/Cohen-Macaulay notion for simplicial complexes that the locally bouquet condition generalizes.","marker":"[Bac80]"},{"why":"Provides the homotopy path algebra setting and the earlier special case of the cellular resolution whose general categorical version is developed here.","marker":"[FH22]"},{"why":"Defines the Reiner-Stamate equivalence relations and reduced incidence algebras whose Koszulity criterion is recovered and reinterpreted.","marker":"[RS10]"},{"why":"Establishes the classical result, with Woodcock, that incidence algebras of locally Cohen-Macaulay graded posets are Koszul, recovered as a corollary here.","marker":"[Pol95]"},{"why":"Gives the complementary classical Cohen-Macaulay-to-Koszul criterion for incidence algebras that the paper recovers from its topological theorem.","marker":"[Woo98]"}],"fun_headline_variants":["Nerve bouquet test for Koszulity of category algebras","Koszulity equals a bouquet condition on the reduced nerve","Topological proof: Koszulity is a nerve bouquet","When is a category algebra Koszul? Check the nerve"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire equivalence assumes the category is indiscretely based: each connected component of the degree-zero morphisms has exactly one morphism between any two objects, so no nontrivial automorphisms exist.","fun_headline_variants_meta":{"raw":{"variants":["Nerve bouquet test for Koszulity of category algebras","Koszulity equals a bouquet condition on the reduced nerve","Topological proof: Koszulity is a nerve bouquet","When is a category algebra Koszul? Check the nerve"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001605,"raw_usage":{"total_tokens":6338,"prompt_tokens":837,"completion_tokens":5501,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":453,"completion_tokens_details":{"reasoning_tokens":5442}},"tokens_in":453,"tokens_out":5501,"duration_ms":34441,"temperature":1.0,"reasoning_tokens":5442,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T04:56:41.529413+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the entrance path category of the two-dimensional torus with the four-strata cube stratification from Example 4.29/4.40; the interval (~v,~w) is two disjoint arcs, so BC(p) for the corresponding path is disconnected and \\tilde $H^{0}$(BC(p)) \\neq 0. The theorem predicts $Ext^{2}$(S_w,S_v)_{-n} \\neq 0 for some n \\neq 2, hence non-Koszulity; a direct computation of $Ext^{2}$ from the quiver with the single cubic relation would settle whether the prediction holds.","supporting_citations":[],"review_version":1}