Maximal Deflation-Exact Structure

• Maximal Deflation-Exact Structure is a categorical framework defining the largest exact structure on additive categories by ensuring deflations remain stable under pullbacks.
• It employs semi-stable kernels and cokernels to form stable pairs, guaranteeing maximal effectiveness in homological algebra, tensor decomposition, and numerical linear analysis.
• This structure unifies classical and modern methods, enabling precise handling of singularities and efficient deflation techniques in various applied mathematical contexts.

A maximal deflation-exact structure is a categorical and algebraic concept that designates the largest possible exact structure on an additive category in which deflations (admissible epimorphisms) are precisely those morphisms stable under pullback. This framework systematically captures situations in homological algebra, tensor decompositions, numerical linear analysis, and singularity resolution in polynomial systems, where maximal sets of “exact” decompositions or deflations are defined, generalized, or exploited.

1. Exact Structures and Deflation-Exact Categories

An additive category $\mathcal{A}$ is a category where morphism sets form abelian groups, composition is bilinear, there is a zero object, and biproducts of finite families exist. The presence of kernels and cokernels for every morphism is assumed. Within this setting, a kernel–cokernel pair (or short sequence) is a diagram

$X \xrightarrow{f} Y \xrightarrow{g} Z$

with $f = \ker(g)$ and $g = \coker(f)$. A class $\mathcal{E}$ of such pairs—closed under isomorphisms, whose first and second maps are called inflations and deflations, respectively—furnishes the data for an exact structure if the axioms (E0), (E1), (E2), and their $\operatorname{op}$ versions are satisfied.

A deflation-exact structure is an exact structure in which the set of deflations comprises precisely those morphisms $g$ that remain deflations under any pullback. Dually, one defines inflation-exact structures. The maximal deflation-exact structure is the largest such structure regarding inclusion on a given additive category (Sieg et al., 2014, Crivei, 2012).

2. Construction: Semi-Stable Kernels/Cokernels and Stable Pairs

The key ingredient is semi-stability. For $g:Y\to Z$, $g$ is a semi-stable cokernel if in every pullback square

$$\begin{CD} P @>p_Y>> Y \ @V p_T VV @VVgV \ T @>t>> Z \end{CD}$$

the morphism $p_T$ is again a cokernel. The kernel notion is dualized via pushouts. The set of all kernel–cokernel pairs for which both maps are semi-stable—termed stable pairs—forms

$\mathcal{E}_{\max} = \{ (f:X\to Y, g:Y\to Z) \mid f = \ker(g),\, g = \coker(f),\, f,g\ \text{semi-stable} \}.$

This collection satisfies the Quillen axioms—stability under compositions and under pushout/pullback operations—and is maximal: every other exact structure $\mathcal{E}'$ must have its deflations (inflations) semi-stable, so $\mathcal{E}'\subseteq\mathcal{E}_{\max}$ (Sieg et al., 2014).

3. Maximality and Characterizations

Rump’s theorem ensures that every additive category admits a unique maximal Quillen exact structure (Crivei, 2012). This structure often coincides with the class of stable short exact sequences—a short sequence is stable if its kernel is semi-stable under pushout and its cokernel under pullback. In abelian and quasi-abelian categories, all kernels and cokernels are automatically semi-stable, so the maximal structure recovers the classical exact structure.

For general additive categories, the maximal deflation-exact structure coincides with the stable pairs if and only if the category is stable under pushouts and pullbacks of these conflations, often equivalently characterized using idempotent completion functors (Crivei, 2012).

Table: Criteria for Maximal Deflation-Exact Structure

Category Type	All Pairs Stable?	Maximal Structure Coincidence
Abelian, quasi-abelian	Yes	Classical (all pairs)
Preabelian, WIC	Yes	Stable pairs (Sieg–Wegner–Crivei)
General additive	Sometimes, via completion	If stable under pushouts/pullbacks

4. Applications: Functional Analysis, Homological Algebra, Tensors, and Numerical Algorithms

In functional analysis, categories such as BOR (bornological locally convex spaces) and HD-BOR (Hausdorff bornological spaces) are not quasi-abelian but are additive with kernels and cokernels. Deflations in $\mathcal{E}_{\max}$ for BOR correspond to continuous surjections whose pullbacks along arbitrary maps remain cokernels; stable pairs furnish a unique maximal exact structure, allowing homological algebra in non-quasi-abelian settings (Sieg et al., 2014).

In tensor decomposition—specifically, the CANDECOMP/PARAFAC (CPD) context—the maximal deflation-exact structure framework refers to the sequential extraction of rank-1 or block (rank-2) terms from a tensor. Here, under suitable conditions (e.g., at least two factor matrices full column rank), each deflation step recovers one or more components exactly, with guaranteed reduction in rank and explicit, tight computational complexity bounds. This process is maximal in that extraction continues until no further exact deflation is possible (Phan et al., 2015).

In numerical linear algebra, for GMRES applied to large, sparse, non-Hermitian systems, the deliberate construction of a deflation subspace using spectral projectors enables the exact removal (“deflation”) of challenging spectral components in one step, leading to a maximal guaranteed reduction in the residual norm at each iteration. The “maximal” deflation space in this context annihilates all spectral modes above a given threshold, and the convergence factor bound is sharp for the remainder (Spillane et al., 2023).

For singular solutions of polynomial systems, maximal (or exact) deflation refers to methods that reduce the multiplicity of an isolated singular root to one, either iteratively (first-order incremental deflation) or all at once (inverse system/dual basis expansion). These constructions ensure maximal descent (the isolated root becomes simple) and complete characterization of the local algebra structure, with polynomial complexity for the direct construction (Hauenstein et al., 2016).

5. Fundamental Theorems and Propositions

Verification of Maximal Structure: The axioms for exact structures ([E0], [E1], [E2], and their duals) are verified for $\mathcal{E}_{\max}$ by showing that identities are semi-stable, semi-stability is preserved under composition (Kelly’s stability), and stability under pushout/pullback is automatic for the defined classes of inflations and deflations (Sieg et al., 2014).

Uniqueness and Pullback/Pushout Behavior: Rump’s theorem guarantees uniqueness; Crivei’s equivalence via the idempotent completion functor $H: \mathcal{A} \to \mathcal{A}^\sharp$ clarifies exactness in arbitrary additive categories (Crivei, 2012).

Maximal Deflation in Tensor Decomposition: When at least two mode factor matrices are full column rank, block deflations (rank-2, or higher in general) succeed at each step, with cost $O(R^K)$ per step for block size $K$, giving a maximal sequence of exact deflations constrained only by the decreasing multilinear rank (Phan et al., 2015).

Maximal Deflation Subspace in GMRES: By building the deflation subspace from generalized eigenvectors tied to the skew–Hermitian and Hermitian parts of the matrix, all selected spectral components are deflated in one iteration, and the resulting convergence factor is explicitly maximal under the established bounds (Spillane et al., 2023).

Polynomial System Singularities: The direct inverse-system construction yields via a polynomial-size system a unique, simple root encoding both the root and its full multiplicity structure; each variable and equation introduced tracks the commutator structure of multiplication matrices or dual basis elements (Hauenstein et al., 2016).

6. Examples and Specializations

• Bornological Spaces: In $\BOR$, the maximal deflation-exact structure is strictly smaller than the naïve pairwise structure due to failure of cokernel semi-stability for certain surjections. The maximal exact structure permits the deployment of derived and triangulated category techniques in these non-quasi-abelian settings (Sieg et al., 2014).
• Free Modules: In the category of free modules over a ring, all split short exact sequences remain free under pushouts and pullbacks, so the class of stable short exact sequences recovers the maximal exact structure (Crivei, 2012).
• Tensor Analytics: In CPD, both ASU-1 (rank-1 deflation) and ASU-2 (rank-2/block deflation) yield guaranteed reduction in rank and exact recovery of extracted components, outperforming classical ALS in complexity per deflation step (Phan et al., 2015).
• Polynomial System Deflation: The iterative or direct approach can be selected based on system size and multiplicity, offering a tradeoff between incremental linear complexity and direct polynomial complexity for capturing isolated singular solution structure (Hauenstein et al., 2016).

7. Implications and Broader Context

The maximal deflation-exact structure systematically extends classical exactness to broad, potentially non-quasi-abelian frameworks and provides a categorical foundation for stable, exact reductions in algebraic, analytic, and algorithmic settings. In categories not admitting all kernels and cokernels as stable, these structures are essential for defining derived categories and facilitating functional-analytic or homological techniques.

In algorithmic applications, maximal deflation-exact constructs enable structured, stepwise reduction of problems: in tensors, by successive exact decomposition; in GMRES, by annihilating troublesome spectral subspaces; and in singularities, by efficiently regularizing roots. These principles unify categorical, analytic, and algebraic approaches to “maximal” and “exact” decomposition, underpinning both theoretical advances and computational methodologies across modern mathematics and applied science (Sieg et al., 2014, Phan et al., 2015, Crivei, 2012, Spillane et al., 2023, Hauenstein et al., 2016).