Connectedness in Codimension One and the Non-$S_2$ Locus

Source: arXiv:2605.06617 · Published 2026-05-07 · By Likun Xie

TL;DR

This paper addresses a structural question in commutative algebra: to what extent do finite S2-objects (coherent S2-sheaves on Noetherian schemes, finitely generated graded S2-modules) decompose according to the topology of their support, specifically the notion of connectedness in codimension 1. The central thesis is that such objects decompose canonically into summands indexed by the connected components in codimension 1 of their support, and this decomposition is essentially unique under mild local indecomposability hypotheses. This generalizes and reframes classical results of Hochster–Huneke on complete local rings, which relate connectedness in codimension 1 of Spec A, indecomposability of the canonical module ωA, and localness of the S2-ification EndA(ωA), by identifying all three as consequences of a single structural principle about S2-objects.

The paper then applies this decomposition framework to the non-S2 locus of a local ring. Given the S2-ification sequence 0 → A → EndA(ωA) → C → 0, the cokernel C has support equal to the non-S2 locus. The key new result (Theorem 1.4) identifies the connecting morphism φ: K^{t+1}(A) → K^t(C) (between deficiency modules of A and C, where t = dim C) as the S2-hull map, so K^t(C) is the S2-hull of K^{t+1}(A). This allows codimension-1 connectedness of the non-S2 locus to be read off from the deficiency module of A itself, without explicitly knowing C.

As a concrete application, the paper studies codimension-2 lattice ideals I_L in a polynomial ring S = k[x_1,...,x_n], using the explicit 3-step minimal resolution of Peeva–Sturmfels. The deficiency module K^{d-1}(A) (where d = n-2 = dim A) is shown to be Γ-graded indecomposable (Γ = Z^n/L), and under additional hypotheses (IL prime, K^{d-1}(A) equidimensional and S2), the top-dimensional part of the non-S2 locus of A is connected in codimension 1. This applies in particular to non-Cohen–Macaulay toric rings of codimension 2 whose canonical module is Cohen–Macaulay, including simplicial affine semigroup rings.

Key findings

• Theorem 3.4/3.6: Every nonzero coherent S2-sheaf (resp. finitely generated Γ-graded S2-module over a torsion-free graded ring) decomposes canonically into a direct sum indexed by the connected components in codimension 1 of its support; this decomposition is unique up to reordering when each summand is locally indecomposable in codimension ≤1.
• Theorem 3.7 (Theorem 1.3): For a local ring A that is a homomorphic image of a Gorenstein local ring, both the canonical module ωA and the S2-ification B = EndA(ωA) decompose canonically with summands ωi and Bi indexed by the codimension-1 connected components Si of Spec(A/J(A)); each Bi is connected as a ring, and if A is complete local, each Bi is local — extending beyond the complete case the earlier combinatorial count of Sather–Wagstaff and Spiroff [19] via the Hochster–Huneke graph.
• Theorem 1.4: For A a local ring (homomorphic image of Gorenstein) with J(A)=0 and nonzero cokernel C in the S2-ification sequence, the connecting map φ: K^{t+1}(A) → K^t(C) (t = dim C) is the S2-hull map, so K^t(C) ≅ S2-hull of K^{t+1}(A); this reduces the question of codimension-1 connectedness of Supp(C) to an indecomposability question about K^{t+1}(A).
• Theorem 1.5: For a codimension-2 lattice ideal IL minimally generated by at least 4 elements (equivalently, A = S/IL non-Cohen–Macaulay), the deficiency module K^{d-1}(A) is Γ-graded indecomposable (Γ = Z^n/L, d = n-2); if additionally IL is prime and K^{d-1}(A) is equidimensional and S2, then the top-dimensional part of the non-S2 locus of A is connected in codimension 1.
• Corollary 1.6: For non-Cohen–Macaulay toric rings of codimension 2 whose canonical module is Cohen–Macaulay (e.g., simplicial affine semigroup rings by [22, Thm. 6.4]), the non-S2 locus coincides with the non-Cohen–Macaulay locus, and its top-dimensional part is connected in codimension 1.
• Proposition 1.7: For lattice ideals IL of codimension 2 minimally generated by exactly 4 elements with A = S/IL equidimensional and unmixed, the non-S2 locus is nonempty of dimension dim A − 2 and its top-dimensional part is connected in codimension 1; this is proved by a direct (non-graded) indecomposability argument applicable even when IL is not prime.
• The S2-extension lemma (Lemma 3.1, attributed to [23, Cor. 1.18]) is the technical engine: for an S2-sheaf F and a closed Y with codim(Y ∩ Supp F) ≥ 2, the natural map F → j_*(F|_U) is an isomorphism (proved algebraically via vanishing of H^0_I(M) and H^1_I(M) under depth ≥ 2); this underpins all decomposition theorems.

Methodology — deep read

The paper is theoretical/algebraic rather than empirical, so 'methodology' here refers to the logical architecture of the proofs and the mathematical setup.

1. Foundational setup and objects of study. The paper works throughout with Noetherian schemes X and coherent sheaves, and with Noetherian Γ-graded rings (Γ torsion-free abelian) and finitely generated graded modules. The key property is Serre's condition S2: a module M over a local ring A satisfies S2 if depth(M_p) ≥ min(2, dim M_p) for all primes p. The paper also works with local rings that are homomorphic images of Gorenstein local rings, which is equivalent (by [10, Cor. 1.4]) to admitting a dualizing complex; this ensures the existence of canonical and deficiency modules K^i(M) := Ext^{n-i}_R(M,R) where R is an ambient Gorenstein ring of dimension n.

2. Core technical lemma (Lemma 3.1, S2-extension property). The crucial input is that an S2-sheaf F on X extends uniquely from U = X \ Y to all of X whenever codim(Y ∩ Supp F) ≥ 2: the map F → j_*(F|_U) is an isomorphism. The algebraic proof proceeds by reducing to an affine module M satisfying S2, using depth ≥ 2 (from the S2 condition and ht I ≥ 2) to conclude H^0_I(M) = H^1_I(M) = 0, and then invoking the ideal-transform exact sequence 0 → H^0_I(M) → M → D_I(M) → H^1_I(M) → 0 to obtain M ≅ D_I(M) ≅ Γ(U, M~).

3. Indecomposability ↔ connectedness (Theorems 3.2 and 3.3). For the forward direction (indecomposable ⟹ connected in codimension 1): if Supp F is not connected in codimension 1, one finds a closed Y of codimension ≥ 2 such that U = Supp F \ Y disconnects into U_1 ⊔ U_2; then F|_U decomposes and by the extension lemma so does F, contradicting indecomposability. For the converse (connected in codimension 1 + local indecomposability in codimension ≤1 ⟹ indecomposable): if F = F_1 ⊕ F_2, the intersection S_1 ∩ S_2 of supports must have codimension ≥ 2 (by local indecomposability), which disconnects Supp F, contradicting the connectivity hypothesis. The graded version (Theorem 3.3) additionally requires the torsion-free hypothesis on Γ to ensure associated primes of graded modules are homogeneous, so that the disconnection argument can be carried out in the Γ-homogeneous setting.

4. Canonical decomposition theorem (Theorems 3.4 and 3.6). Given the decomposition Supp F = S_1 ∪ … ∪ S_r into codimension-1 connected components, the pairwise intersections S_i ∩ S_j have codimension ≥ 2. Setting Y = ∪{i≠j}(S_i ∩ S_j) and U = X \ Y, the restriction F|U decomposes as ⊕ F|. The extension lemma then pushes this back to a global decomposition F ≅ ⊕ F^(i) with F^(i) = (j_i)*(F|_{U_i}). Canonicality follows because U_i is determined by S_i. Uniqueness of the indecomposable decomposition is proved by showing any indecomposable summand G_j must have Supp G_j ⊆ S_i for some i (using Theorem 3.2), then the extension lemma forces G_j to coincide with F^(i) after aggregation.

5. Application to canonical modules and S2-ifications (Theorem 3.7). For a local ring A (homomorphic image of Gorenstein), ωA is equidimensional S2 with Supp(ωA) = Spec(A/J(A)), so Theorem 3.4 applies directly to decompose ωA. Local indecomposability at codimension-≤1 points follows because A_p ≅ End_{A_p}((ωA)_p) for such p (the ring equals its own S2-ification at codimension ≤1 primes). The decomposition of B = EndA(ωA) into ∏ B_i follows because HomA(ωi, ωj) = 0 for i ≠ j (their supports are disjoint in codimension 1, so no associated primes of ωj lie in Supp(ωi)). Part (c) uses a snake lemma diagram with the diagonal map A/J(A) → ⊕ A_i (whose cokernel has dimension ≤ dim A − 2) to relate the global cokernel C to the individual cokernels C_i.

6. Deficiency module and S2-hull (Theorem 1.4). The S2-ification sequence 0 → A → B → C → 0 induces a long exact sequence in local cohomology, giving a connecting map φ: K^{t+1}(A) → K^t(C) (t = dim C). The paper identifies φ as the S2-hull map by checking the conditions of [12, Def. 9.3]: φ is an isomorphism in codimension ≤0 within Supp(C) (since Supp(C) = non-S2(A) has height ≥ 2 in Spec A, and B is S2 so B_p ≅ A_p at height-1 primes), and K^t(C) is S2 (by Proposition 2.1(b)). The conclusion K^t(C) ≅ S2-hull of K^{t+1}(A) then allows indecomposability of K^{t+1}(A) (which is easier to compute) to imply indecomposability of K^t(C), which in turn implies connectedness in codimension 1 of Supp(C) = non-S2(A).

7. Lattice ideal application (Theorems 1.5 and Proposition 1.7). For a codimension-2 lattice ideal IL ⊆ S = k[x_1,...,x_n], the 3-term minimal free resolution of Peeva–Sturmfels [18] (of the form 0 → F_2 → F_1 → F_0 → IL → 0, with explicit Γ-graded Betti numbers) is used to compute K^{d-1}(A) via Ext. The graded indecomposability of K^{d-1}(A) is established by showing it has a unique minimal prime in each Γ-graded component, using the explicit combinatorial structure of the resolution. For the non-prime case with exactly 4 minimal generators, a direct module-theoretic argument (not relying on the graded structure) establishes ordinary indecomposability. The paper is transparent that the step requiring K^{d-1}(A) to be equidimensional and S2 is an additional hypothesis whose verification in individual cases may require separate work; it provides the simplicial affine semigroup ring case as one where this can be confirmed via [22, Thm. 6.4] and [2, Prop. 2.2].

Technical innovations

• The S2-extension lemma is used as a universal tool to reduce global decomposition problems for S2-objects to purely topological (codimension-1 connectivity) questions about the support, replacing the earlier complete-local and combinatorial (Hochster–Huneke graph) arguments of [9] and [19] with a scheme-theoretic argument that works without completeness.
• Theorem 1.4 identifies the connecting map φ: K^{t+1}(A) → K^t(C) in the long exact deficiency sequence as the S2-hull map, giving a new and explicit relationship between deficiency modules of a ring and of its defect locus; prior work (Schenzel [24]) studied ideal transforms and local cohomology in this context but did not characterize this map as an S2-hull.
• The graded decomposition theorem (Theorem 3.6) for Γ-graded S2-modules over torsion-free graded rings is new and provides an algebraic analogue of the sheaf decomposition, carefully handling the subtlety that the torsion-free hypothesis on Γ is necessary for associated primes of graded modules to be homogeneous.
• The application to codimension-2 lattice ideals (Theorem 1.5) combines the Peeva–Sturmfels resolution [18] with the S2-hull/deficiency module framework to obtain connectedness-in-codimension-1 results for the non-S2 locus without requiring the ring to be Cohen–Macaulay or the ideal to be prime, extending beyond what was previously accessible via the Hochster–Huneke framework.

Limitations

• The main connectedness result for codimension-2 lattice ideals (Theorem 1.5) requires the additional hypotheses that IL is prime and that K^{d-1}(A) is equidimensional and S2; the paper acknowledges these are not always automatic and provides only specific families (simplicial affine semigroup rings, CM canonical module case) where they can be verified.
• The converse direction of the indecomposability–connectedness equivalence (Theorem 3.2) requires a local indecomposability hypothesis at all primes of codimension ≤1 in Supp F; the paper notes (Remark 3.5) that without equidimensionality this condition would need to be formulated componentwise, and examples showing the necessity of these conditions are not systematically provided.
• The paper works exclusively with rings that are homomorphic images of Gorenstein rings (equivalently, admit dualizing complexes); the theory of deficiency modules and canonical modules used throughout is not available in full generality beyond this class, limiting direct applicability to, e.g., non-excellent rings or more exotic Noetherian rings.
• The non-prime lattice ideal case (Proposition 1.7, exactly 4 generators) is handled by a direct argument specific to that generator count; the paper does not provide a uniform treatment for non-prime IL with more than 4 generators, and explicitly notes the graded indecomposability argument breaks down when the grading group Γ = Z^n/L has torsion.
• No computational examples are worked through in full detail for large or complicated rings; the illustrative examples are confirmatory rather than exploratory, and there is no systematic comparison with prior connectedness results (e.g., Faltings' theorem, Grothendieck's connectedness theorem) in terms of which gives stronger conclusions in overlapping cases.

Open questions / follow-ons

• Can the equidimensionality and S2 hypotheses on K^{d-1}(A) in Theorem 1.5 be removed or replaced by weaker ring-theoretic conditions that are more readily checkable for arbitrary codimension-2 lattice ideals, including the non-prime case with more than 4 generators?
• The paper establishes that the non-S2 locus of suitable codimension-2 toric rings is connected in codimension 1; do analogous results hold for the non-S_r loci (r ≥ 3) or for Macaulayfication defect loci, using higher deficiency modules in place of K^{d-1}(A)?
• The decomposition theory is developed for S2-sheaves and S2-modules; can an analogous canonical decomposition theorem be formulated for S_r-objects (r ≥ 3) using connectedness in codimension r, and what is the correct analogue of the S2-extension lemma in that setting?
• The graded version (Theorem 3.6) requires Γ to be torsion-free; is there a natural analogue or obstruction theory for torsion grading groups, e.g., for lattice ideals where the quotient group Z^n/L has torsion, and can the non-S2 connectedness results be extended to that setting?

Why it matters for bot defense

This paper is pure commutative algebra and algebraic geometry with no connection to CAPTCHA systems, bot defense, machine learning, or any aspect of web security or behavioral analysis. Its subject matter — S2-sheaves, canonical modules, deficiency modules, lattice ideals, connectedness in codimension 1 — is entirely within abstract algebra and has no applied relevance to bot detection, challenge-response systems, or traffic classification.

A bot-defense or CAPTCHA engineer would find no applicable techniques, threat models, datasets, or algorithmic ideas in this work. It should not be prioritized for reading by practitioners in that domain.

Cite

@article{arxiv2605_06617,
  title={ Connectedness in Codimension One and the Non-$S_2$ Locus },
  author={ Likun Xie },
  journal={arXiv preprint arXiv:2605.06617},
  year={ 2026 },
  url={https://arxiv.org/abs/2605.06617}
}

Read the full paper

• arXiv:2605.06617 (abstract)
• arXiv:2605.06617 (PDF)