Finite-length projective dimensions over the diagonal quotient

Let k be any field, S a finite lattice, A the interval-pair incidence algebra, I its diagonal ideal, B = A/I, and C = Bᵐᵒᵖ. For each strict vertex v, let Vᵥ be the actual right B-module, viewed as a left C-module, constructed in the simple-module classification.

For a finite-length M : ModuleCat C, the library proves

projectiveDimension M ≤ ⨆ v : StrictVertex S, projectiveDimension Vᵥ.

This is an inequality in Mathlib's WithBot ℕ∞: it includes the zero-module value ⊥ and allows an infinite value. It gives no numerical estimate and does not require the supremum to be attained. The public bound interface is also available at each natural degree: if every Vᵥ has HasProjectiveDimensionLE Vᵥ d, every finite-length M has HasProjectiveDimensionLE M d. Conversely, a bound on every finite-length module gives the same bound on every Vᵥ, since each actual vertex module is simple and therefore finite length. These implications form the public uniform-bound equivalence.

The finite-length lifting theorem follows Mathlib's IsFiniteLength induction. In the simple quotient case, it uses the actual row N → M → M/N formed by Submodule.subtype and Submodule.mkQ. The short-exactness proof is the module category's subtype/quotient exactness result. The quotient is simple by the induction constructor, so the #699 classification supplies a C-linear isomorphism to one of the actual Vᵥ. Transporting its dimension bound and applying ShortExact.hasProjectiveDimensionLT_X₂ completes the extension step. The zero case uses Mathlib's zero-object criterion for HasProjectiveDimensionLT.

For the unconditional inequality, write the candidate bound as D = ⨆ v, projectiveDimension Vᵥ. The proof unfolds projectiveDimension as the infimum of bounds n : WithBot ℕ∞ for which every natural degree i : ℕ with n < i satisfies HasProjectiveDimensionLT M i. Whenever D < i, each vertex dimension is below i; projectiveDimension_lt_iff then supplies the common Ext-vanishing degree used in the finite-length theorem. Thus D is admissible for M, even when the supremum is infinite or the vertex set is empty. No degree value is introduced as a premise.

The underlying-field version uses the existing restrictRightQuotientModule M action. Its finite-dimensionality gives both Noetherianity and Artinianity over k. With the actual scalar tower k → C → M, Mathlib transfers these properties to C; the finite-length criterion then gives IsFiniteLength C M. The already established inequality and uniform-bound equivalence therefore apply to every module finite-dimensional over this restricted k-module. Each Vᵥ itself is one-dimensional after restriction by rightVertexModule_restrictKIso, so the reverse uniform-bound implication remains available in this class. No finite-field, algebraic-closure, semisimplicity, or chosen-composition-series premise is added.

This is the finite-length reduction step suggested by BOOS24-v3, §2.4 and the first paragraph of Lemma 5.7, using the actual quotient-simple classification from #699 and Notation 5.9. It does not formalize the paper's subsequent tensor/base-change resolution, prove a quotient-versus-incidence global-dimension inequality, or assert the numerical 2n−2 bound.

The finite-length induction structure adapts ModuleCat.hasProjectiveDimensionLT_of_isFiniteLength from TauCeti's Dimension.lean at commit 7b0d106bf79dbea38ba8819985bd029989181287. That source is copyright © 2026 The Tau Ceti contributors and is licensed under Apache-2.0, as stated in its file header and LICENSE. This module adapts the induction to the quotient-specific vertex bound and does not import or add TauCeti as a dependency. Its public API remains specific to this quotient; it does not add a second generic finite-length framework.