Regular Matroids: Structure & Applications

• Regular matroids are combinatorial structures characterized by total unimodularity and the exclusion of specific minors, enabling versatile representations in optimization and polyhedral theory.
• Seymour’s decomposition theorem shows that every regular matroid can be constructed from graphic, cographic, and R10 matroids using 1-, 2-, and 3-sum operations.
• Algorithmic applications leverage regular matroids to achieve fixed-parameter tractability for complex problems and to derive succinct extended formulations for associated polytopes.

A regular matroid is a matroid representable over every field, equivalently, one that admits a representation by a totally unimodular matrix (every square submatrix has determinant $0$, $1$, or $-1$). Regular matroids encompass important classes such as graphic and cographic matroids and serve as a foundational object in combinatorial optimization, matroid theory, and polyhedral theory, owing to their rich decomposition and representation structure.

1. Definition, Characterizations, and Fundamental Properties

A matroid $M$ is regular if there exists a totally unimodular matrix $A$ such that $M$ is isomorphic to the matroid defined by the column dependencies of $A$. Some additional characterizations and foundational results include:

• Binary Matroids: Every regular matroid is binary, but not conversely; not all binary matroids are regular (Fomin et al., 2016).
• Excluded Minor Characterization: Regular matroids are precisely those binary matroids with no $U_{2,4}$, $F_7$ (Fano), or $F_7^*$ (non-Fano) minor.
• Duality: The dual of a regular matroid is regular (0909.5033).
• Operations: Regularity is preserved under minors, direct sums, 2-sums, and 3-sums (Dvorak et al., 24 Sep 2025).
• Special Classes: Graphic (cycle) matroids, cographic (bond) matroids, and the $R_{10}$ matroid are all regular (Dinitz et al., 2012).

A deep algebraic-geometric perspective expresses regularity via moduli: a matroid is regular if and only if its "foundation" (a universal invariant) is the regular partial field, essentially $\{0,1,-1\}$ with $0=1+(-1)$. This connects regularity to being both binary and orientable (admitting signed representations) (Baker et al., 2018).

2. Structural Theorems: Seymour’s Decomposition

Seymour’s decomposition theorem is central to regular matroid theory. It yields a recursive structure theorem:

Seymour’s Decomposition Theorem: Every regular matroid can be constructed from graphic, cographic, and the 10-element matroid $R_{10}$ via a sequence of 1-sums, 2-sums, and 3-sums (Dinitz et al., 2012, Fomin et al., 2017, Dvorak et al., 24 Sep 2025).

More formally, any regular matroid $M$ can be decomposed as

$$M = (((M_1 \oplus_1 M_2) \oplus_2 M_3) \oplus_3 \cdots \oplus_i M_k), \quad i \in \{1,2,3\}$$

where each $M_j$ is graphic, cographic, or isomorphic to $R_{10}$.

In recent advances, the forward (composition) direction of this theorem has been formally verified in Lean 4, confirming that 1-, 2-, 3-sums of regular matroids (via totally unimodular matrix construction at each step) are regular, even for infinite ground sets with finite rank (Dvorak et al., 24 Sep 2025).

3. Algorithmic and Polyhedral Aspects

The decomposition theorem underpins several algorithmic and polyhedral results for regular matroids:

• Parameterization and Fixed-Parameter Tractability: Many fundamental problems that are intractable on general matroids become FPT on regular matroids when parameterized suitably (e.g., the Space Cover, Spanning Circuit, and Minimum Spanning Circuit problems). Algorithms exploit the decomposition into basic pieces and dynamic programming over the decomposition tree (Fomin et al., 2016, Fomin et al., 2017).

Example: On a regular matroid $M$, the Space Cover problem is FPT in parameter $k$ (the sought cover size) with runtime $20^k \cdot |M|^{O(1)}$, combining branching and bottom-up composition over the decomposition (Fomin et al., 2017).

• Matroid Secretary and Online Selection Problems: A $9e$-competitive (constant-competitive) matroid secretary algorithm leverages Seymour decomposition, coordinating decision-making across graphic, cographic, and $R_{10}$ leaves with synchronized independence via contraction and deletion of shared elements (Dinitz et al., 2012).
• Extension Complexity: The independence polytope of a regular matroid on $n$ elements has extension complexity $O(n^6)$, with circuit dominant polytopes enjoying $O(n^2)$ extended formulations. The decomposition into basic matroids allows inductive construction of small extended formulations despite the complexity of the 3-sum operation (Aprile et al., 2019).

4. Circuit Structure and Combinatorial Invariants

Regular matroids exhibit controlled circuit behavior, useful in both combinatorics and optimization:

• Circuit Counting: For any regular matroid on $m$ elements and $\alpha \geq 1$, the number of circuits of size at most $\alpha$ times the minimum circuit size is polynomially bounded, $m^{O(\alpha^2)}$, generalizing Karger's result for cuts in graphs and extending to codes and lattice vector enumeration (Gurjar et al., 2018).
• Circuit-Difference Structure: In regular matroids, the class of circuit-difference matroids (where symmetric difference of intersecting circuits is again a circuit) admits an explicit characterization: such a matroid has no pair of skew circuits (those with $r(X\cup Y)=r(X)+r(Y)$), and the class is closed under series minor operations (Drummond et al., 2020).
• Cocircuit and Graphic Matroids: The concept of “graphic cocircuit” is introduced: a cocircuit $Y$ is graphic if $M\setminus Y$ is graphic. A regular matroid with all graphic cocircuits is signed-graphic if and only if it lacks two specific minors: $M^*(G_{17})\cong M^*(K_{3,5})$ and $M^*(G_{19})\cong M^*(K_{4,4}$ minus an edge) (0909.5033).

Class	Definition	Construction/Characterization
Regular	Representable over $\forall$ fields; TU matrix exists	Seymour decomposition: sums of graphic/cographic/$R_{10}$
Circuit-diff.	Intersecting circuits: $\Delta$ always a circuit	No pair of skew circuits
Signed-graphic	Graphic cocircuits, forbidden minors absent	Lacks $M^*(K_{3,5})$, $M^*(K_{4,4}-e)$ as minors

5. Connections to Algebra, Geometry, and Invariant Theory

Algebraic geometry and invariant theory provide new lenses on regular matroids:

• Moduli Spaces and "Foundations": The moduli space $Mat(r,E)$ of rank-$r$ matroids frames each classical matroid as a point over the Krasner hyperfield, with the "foundation" $k_M^f$ invariant distinguishing regularity: $M$ is regular iff $k_M^f$ is the regular partial field. Regularity coincides with “binary and orientable” in this moduli view (Baker et al., 2018).
• Dressians and Polyhedral Subdivisions: Dressians parametrize matroid subdivisions of the matroid polytope: $\mathcal{D}r_M$ is the tropical prevariety recording all weightings $w$ on the bases inducing matroidal subdivisions. Regular matroids fit naturally within this tropical framework, and computational tools (elimination of extraneous linearity) can reduce computation complexity (Brandt et al., 2019).
• Symmetric Edge Polytopes: For any regular matroid (via a full-rank weakly unimodular representation), one associates a symmetric edge polytope $P_M = \operatorname{conv}\{M, -M\}$, generalizing classical constructions for graphs. The Ehrhart theory of the polar $P_M^\Delta$ is intimately linked to the lattice of flows of the dual regular matroid, with facets and triangulations dictated by circuit/cut structure (D'Alì et al., 2023).

6. Applications, Algorithmic and Polyhedral Implications

Regular matroids' structure enables numerous applications:

• Tutte Polynomials and Generalizations: The algorithmic computation of the Tutte polynomial for small regular matroids is feasible via internal and external base activity (Fripertinger et al., 2011). Generalizations (the $A$-polynomial) extend the Tutte invariant to regular oriented matroids, encoding information such as acyclicity and total cyclicity (Awan et al., 2022).
• Sandpile Groups and Torsor Algorithms: The sandpile group $S(M) = \mathbb{Z}^E/(\Lambda(M)\oplus\Lambda^*(M))$ acts simply transitively on the set of bases. The existence of canonical, “consistent” sandpile torsor actions (using BBY bijections and triangulating signatures) has been affirmatively established for regular matroids, with compatibility under deletion and contraction (Ding et al., 4 Jul 2024). For Jacobians, the cardinality and structure, including the exponent 2 classification, extends from graphs to regular matroids (Lheem et al., 2019).
• Matroid Splitting Operations: The effect of the splitting operation is tightly controlled; only regular matroids avoiding certain forbidden minors maintain graphicness after splitting (Mundhe et al., 2020).

7. Further Directions and Open Problems

Several research directions and unresolved questions remain:

• Decomposition Formalization: While the forward direction of Seymour's theorem is now formally verified (Dvorak et al., 24 Sep 2025), the decomposition (reverse) direction for infinite matroids or without finiteness assumptions remains open.
• Algorithmic Gaps: While FPT results are achievable for regular matroids, extensions to broader classes of binary matroids are hindered by W[1]-hardness even for closely related problems (Fomin et al., 2016). For cographic matroids, some parameterizations also lead to hardness.
• Optimization and Polyhedral Theory: Tightening extension complexity bounds remains an open problem—specifically, whether an additive bound for 3-sum remains possible in all cases (Aprile et al., 2019). The structure of regular matroids places them in a unique position between efficient polynomial-time optimization (due to total unimodularity) and complex combinatorial behavior.
• Algebraic Characterizations: Exploration of the "irregularity parameter" $p(M)$ for orientable matroids, monotonicity under minors, and the search for forbidden minor characterizations for "almost regular" classes are ongoing (Taylor, 2017).

---

Regular matroids thus serve as a unifying class in matroid theory, combining strong representation-theoretic, structural, algorithmic, and polyhedral properties. Their decomposition, algebraic invariants, and circuit structure continue to stimulate advances in both theory and applications across combinatorics, optimization, geometry, and algebra.