Signed Graphic Matroid Overview

• Signed graphic matroids are defined as those isomorphic to the frame matroid of a signed graph, characterized by circuits such as single positive cycles and various unbalanced cycle combinations.
• They utilize a flow–tension framework where minimal supports of nonzero flows and tensions provide a linear-algebraic perspective that clarifies representability over fields like GF(3).
• Their study advances excluded minor characterizations, recognition algorithms, and connections with graphic, bicircular, and gain-signed matroid constructions.

A signed-graphic matroid is a matroid isomorphic to the frame matroid $M(\Sigma)$ of a signed graph $\Sigma=(G,\sigma)$, where $G=(V,E)$ is a graph and $\sigma:E\to\{+1,-1\}$ assigns a sign to each edge. Signed-graphic matroids arise naturally as a common generalization of graphic matroids and of certain transversal constructions, and they play a key role in the study of matroids representable over fields of characteristic $\ne 2$ (Sivaraman, 2013). In the literature summarized here they also appear as frame matroids of signed graphs and, in biased-graph language, as Zaslavsky frame- or bias-matroids (Papalamprou et al., 2012).

1. Definitions and circuit structure

A signed graph is a pair $\Sigma=(G,\sigma)$ with $G=(V,E)$ and $\sigma:E\to\{+1,-1\}$. A cycle, or circle, is positive when the product of its edge-signs is $+1$, and negative otherwise. The frame matroid $M(\Sigma)$ has ground set $\Sigma=(G,\sigma)$0; a set $\Sigma=(G,\sigma)$1 is independent exactly when, in each connected component of the signed subgraph $\Sigma=(G,\sigma)$2, there is no positive circle and at most one negative circle. A matroid is signed-graphic if it is isomorphic to $\Sigma=(G,\sigma)$3 for some signed graph (Sivaraman, 2013).

For ordinary signed graphs, one standard circuit description lists exactly three types of circuits: a single positive cycle, the union of two negative cycles sharing exactly one vertex, and the union of two vertex-disjoint negative cycles together with a simple path joining them (Lee, 2014). In the frame-matroid formulation via biased graphs, the circuit list is given as minimal edge-sets of a balanced cycle, two unbalanced cycles sharing exactly one vertex, two vertex-disjoint unbalanced cycles joined by a path, or a theta-graph all of whose three cycles are unbalanced (McGuinness, 2019). These descriptions are the basic combinatorial signatures of signed-graphic dependence.

Several papers also work with enlarged graph models. Half-edges are always taken to be negative and loose edges positive; in that setting negative loops and half-lines contribute additional one-dimensional unbalanced phenomena in the signed graph, and the resulting matroidal language still uses the frame-matroid formalism (0909.5033).

2. Rank, cocircuits, and the flow–tension viewpoint

A recent foundational formulation constructs the frame matroid from chain groups. For a signed graph $\Sigma=(G,\sigma)$4, one defines the real flow space

$\Sigma=(G,\sigma)$5

The circuits of the frame matroid are then precisely the minimal supports of nonzero flows in $\Sigma=(G,\sigma)$6, while the cocircuits are the minimal supports of nonzero tensions in

$\Sigma=(G,\sigma)$7

In this formulation the frame matroid emerges from Tutte’s construction of a matroid from the minimal supports of nonzero vectors in a real subspace, and bonds appear as the cocircuits of the signed-graphic matroid (Chen, 15 Aug 2025).

The same framework gives a rank formula. If $\Sigma=(G,\sigma)$8 denotes the number of balanced components of the signed subgraph on $\Sigma=(G,\sigma)$9, then

$G=(V,E)$0

In particular, when $G=(V,E)$1 is an ordinary graph with all edges positive, the flow space is the usual cycle space of $G=(V,E)$2, the minimal supports are the simple cycles, and one recovers the ordinary graphic matroid (Chen, 15 Aug 2025).

This flow–tension perspective is significant because it replaces an ad hoc catalogue of circuit topologies by a linear-algebraic definition. A plausible implication is that it clarifies why signed-graphic matroids fit naturally into representability theory: the circuit and cocircuit systems are simultaneously visible as minimal supports in orthogonal subspaces.

3. Matrix representations and equivalence of representations

Signed-graphic matroids have a natural representation over $G=(V,E)$3. If $G=(V,E)$4 is oriented in the usual bidirected way—positive edges with one head and one tail, negative edges with both ends toward their incident vertices or both away—then the signed incidence matrix $G=(V,E)$5 over $G=(V,E)$6 has entries

$G=(V,E)$7

The column matroid $G=(V,E)$8 is exactly $G=(V,E)$9 (Lee, 2014). This is the representational basis for the implication “signed-graphic $\sigma:E\to\{+1,-1\}$0 ternary” that appears in later structural results.

Projective equivalence of matrix representations has a particularly concrete interpretation in this setting. Two matrices $\sigma:E\to\{+1,-1\}$1 over a field $\sigma:E\to\{+1,-1\}$2 represent the same matroid up to column relabeling precisely when

$\sigma:E\to\{+1,-1\}$3

where $\sigma:E\to\{+1,-1\}$4 is an invertible row-scaling diagonal matrix and $\sigma:E\to\{+1,-1\}$5 an invertible column-scaling diagonal matrix. For signed-graphic representations, row scaling corresponds to resigning all edges in a vertex-cut, while column scaling is also matroid-preserving. Lee shows that row operations alone do not always suffice: there are signed-graphic representations of the same matroid that are not row-equivalent, but become equivalent after a nontrivial column-scale (Lee, 2014).

The same paper introduces the cylinder flip, a matroid-preserving operation on a signed graph embedded on a cylinder. Under suitable blocking-pair hypotheses, the central band is detached, flipped end-for-end, and re-glued; a circuit-by-circuit argument shows that the same edge-set is a circuit before and after the move, hence the matroid is unchanged. Lee also emphasizes two necessary conditions for signed-graphicness: every signed-graphic matroid is dyadic and binet, and testing binet-ness alone is not sufficient unless one also allows column-scaling (Lee, 2014).

4. Excluded minors, decomposition, and recognition

For regular matroids with graphic cocircuits, Papalamprou and Pitsoulis prove a sharp excluded-minor theorem: $\sigma:E\to\{+1,-1\}$6 where $\sigma:E\to\{+1,-1\}$7 and $\sigma:E\to\{+1,-1\}$8 (0909.5033). In the cographic case they give a polynomial-time recognition algorithm: decompose into 3-connected minors via 1- and 2-sums, reconstruct the unique graph for each 3-connected graphic summand, and test membership in the six graph families

$\sigma:E\to\{+1,-1\}$9

In the binary setting with graphic cocircuits, the obstruction list changes. The theorem of Papalamprou and Pitsoulis becomes: $\ne 2$0 (Papalamprou et al., 2012). The proof uses a lifting lemma for cocircuits together with a computer-assisted case analysis showing that the other excluded minors force a cocircuit deletion that is not graphic.

A different structural approach is cocircuit deletion. For a connected binary matroid $\ne 2$1 with a nongraphic cocircuit $\ne 2$2, Papalamprou and Pitsoulis show that $\ne 2$3 is signed-graphic if and only if $\ne 2$4 is bridge-separable and the $\ne 2$5-components are graphic apart from exactly one signed-graphic component (Papalamprou et al., 2010). The theorem is explicitly not a $\ne 2$6-sum decomposition: its organizing object is a cocircuit whose deletion splits the matroid into graphic pieces plus a unique signed-graphic remainder.

This decomposition extends beyond the binary case. For internally 4-connected matroids representable over $\ne 2$7 but not over $\ne 2$8, a nongraphic cocircuit $\ne 2$9 again governs the structure: $\Sigma=(G,\sigma)$0 is signed-graphic with a jointless representation if and only if $\Sigma=(G,\sigma)$1 is bridge-separable and the two contracted minors determined by the avoiding bridge classes are graphic and signed-graphic, respectively (Pitsoulis et al., 2015). These results supply the theoretical basis for recognition algorithms on substantial subclasses of signed-graphic matroids.

5. Relations with graphic and bicircular matroids

Graphic matroids form the most immediate subclass: if all edges of a signed graph are declared positive, then $\Sigma=(G,\sigma)$2 is graphic (Sivaraman, 2013). This inclusion is part of a wider pattern in which signed-graphic matroids interpolate between purely graphic behavior and more general frame-matroid phenomena.

A particularly precise interaction occurs with bicircular matroids. For a graph $\Sigma=(G,\sigma)$3, the bicircular matroid $\Sigma=(G,\sigma)$4 declares $\Sigma=(G,\sigma)$5 independent precisely when each component of $\Sigma=(G,\sigma)$6 contains at most one cycle; its circuits are exactly the minimal edge-sets of subgraphs with two or more cycles, namely subdivisions of a theta-graph, a tight handcuff, or a loose handcuff. Sivaramaran proves that for a graph $\Sigma=(G,\sigma)$7 the following are equivalent: $\Sigma=(G,\sigma)$8 is signed-graphic; $\Sigma=(G,\sigma)$9 is ternary; $G=(V,E)$0 is near-regular; every component of $G=(V,E)$1 can be built, by repeatedly adding pendant edges, from a subdivision of a tree in which some vertices carry loops and some edges are doubled, with tripling or quadrupling allowed only when one or both endpoints is a loopless pendant vertex; and $G=(V,E)$2 has no subgraph homeomorphic to any of six forbidden graphs $G=(V,E)$3. Condition (5) is therefore a purely graph-theoretic forbidden-subdivision characterization of exactly those graphs whose bicircular matroids are signed-graphic (Sivaraman, 2013).

The examples are illustrative. If $G=(V,E)$4 is the six-skein, then $G=(V,E)$5, and since $G=(V,E)$6 is not ternary, $G=(V,E)$7 is not signed-graphic. Conversely, when a graph is built from a looped–doubled tree of the allowed type, one can sign all loops and exactly one edge of each parallel pair negatively so that every cycle is negative and $G=(V,E)$8. The same paper recovers Matthews’s 1977 theorem characterizing when $G=(V,E)$9 is graphic, and records the class implications

$\sigma:E\to\{+1,-1\}$0

while Matthews had shown

$\sigma:E\to\{+1,-1\}$1

6. Generalizations, neighboring constructions, and further directions

The signed-graphic construction extends to gain-signed graphs. An abelian gain-signed graph is a triple

$\sigma:E\to\{+1,-1\}$2

where $\sigma:E\to\{+1,-1\}$3 is a signed graph and $\sigma:E\to\{+1,-1\}$4 is a gain function into an abelian group satisfying $\sigma:E\to\{+1,-1\}$5. For gains in the additive group of a field $\sigma:E\to\{+1,-1\}$6 with $\sigma:E\to\{+1,-1\}$7, the associated matroid has rank

$\sigma:E\to\{+1,-1\}$8

and an extended incidence matrix in $\sigma:E\to\{+1,-1\}$9 represents the one-point extension $+1$0. Deleting the top gain row recovers Zaslavsky’s classical signed-graph frame representation (Anderson et al., 2022). This places signed-graphic matroids inside a larger gain-parametrized family with both vector and affinographic hyperplane representations.

A related but distinct construction is the even cycle matroid. Given a signed graph $+1$1, the even cycle matroid $+1$2 is the binary matroid whose circuits are the $+1$3-even cycles. Its rank is

$+1$4

where $+1$5 if $+1$6 contains an odd cycle and $+1$7 otherwise. Whitney-flips, signature exchanges, and Lovász-flips preserve even cycle matroids (Guenin et al., 2011). This distinction matters: not every matroid naturally attached to a signed graph is a signed-graphic matroid.

Signed-graphic matroids also enter commutative algebra. For any signed graph $+1$8, the toric ideal $+1$9 is generated by quadratic binomials arising from single symmetric exchanges of bases. In the terminology of the paper, signed graphic matroids are linear frame matroids, and White’s Conjecture holds for them (McGuinness, 2019).

Several open directions remain explicit in the literature. Problems singled out include characterizing which signed-graphic matroids are themselves bicircular, extending the forbidden-minor approach to other classes such as transversal matroids, and exploring the algorithmic complexity of testing bicircularity together with signed-graphic realizability (Sivaraman, 2013). These questions reflect the broader position of signed-graphic matroids at the intersection of graph-theoretic cycle structure, matroid representation theory, and the combinatorics of signed and gain-labelled networks.