Whiteley’s Cofactor Matroid in Rigidity

• Whiteley’s Cofactor Matroid is a structure defined on the edge set of complete graphs via the row-dependence of cofactor matrices in bivariate spline theory.
• It bridges spline theory, rigidity, and matroid theory by characterizing 3-dimensional rigidity through maximal K5 circuits and combinatorial rank formulations.
• The matroid underpins advanced graph reconstruction and connectivity analyses, employing operations that preserve independence in flexible circuit configurations.

Whiteley’s cofactor matroid is a matroid on the edge set of a complete graph obtained from the row-dependence structure of a cofactor matrix arising in bivariate spline theory. For generic planar placements, the generic $C^{d-2}_{d-1}$-cofactor matroid $\mathcal{C}_{d-1,n}^{d-2}$ is the row matroid of the corresponding cofactor matrix on $K_n$; in dimension three this is the generic $C_2^1$-cofactor matroid $\mathcal{C}_{2,n}^1$, which is central in the theory of abstract $3$-rigidity matroids (Clinch et al., 2019). On special configurations it coincides with bar-and-joint rigidity on the moment curve and with hyperconnectivity on monomial vectors, placing it at the intersection of spline theory, rigidity theory, and matroid theory (Ruiz et al., 2021).

1. Matrix construction and generic definition

Let $G=(V,E)$ be a finite simple graph, let $p:V\to\mathbb{R}^2$ be a placement with $p(v_i)=(x_i,y_i)$, and let $s\ge 0$. Whiteley’s $\mathcal{C}_{d-1,n}^{d-2}$0-cofactor matrix $\mathcal{C}_{d-1,n}^{d-2}$1 is the $\mathcal{C}_{d-1,n}^{d-2}$2 matrix whose rows are indexed by edges and whose columns come in blocks of $\mathcal{C}_{d-1,n}^{d-2}$3 associated to the vertices. For an edge $\mathcal{C}_{d-1,n}^{d-2}$4 with $\mathcal{C}_{d-1,n}^{d-2}$5, the row has the form

$\mathcal{C}_{d-1,n}^{d-2}$6

where

$\mathcal{C}_{d-1,n}^{d-2}$7

The generic $\mathcal{C}_{d-1,n}^{d-2}$8-cofactor matroid $\mathcal{C}_{d-1,n}^{d-2}$9 is defined as the row matroid of $K_n$0 for a generic map $K_n$1, meaning that the coordinates are algebraically independent over $K_n$2. For $K_n$3-dimensional rigidity, Whiteley’s cofactor matroid is the case $K_n$4, namely $K_n$5 (Clinch et al., 2019).

A closely related notation, used for planar point configurations $K_n$6, starts from the cofactor vector

$K_n$7

The degree-$K_n$8 cofactor rigidity matrix $K_n$9 is then the $C_2^1$0 block matrix whose row $C_2^1$1 has block $C_2^1$2 in position $C_2^1$3, block $C_2^1$4 in position $C_2^1$5, and zeros elsewhere. The associated cofactor rigidity matroid $C_2^1$6 is the linear matroid of the rows of $C_2^1$7, with ground set $C_2^1$8 (Ruiz et al., 2021).

In the three-dimensional case, the cofactor vector reduces to

$C_2^1$9

For a generic framework $\mathcal{C}_{2,n}^1$0, the right kernel of $\mathcal{C}_{2,n}^1$1 contains six independent trivial $\mathcal{C}_{2,n}^1$2-motions. A $\mathcal{C}_{2,n}^1$3-motion is a map $\mathcal{C}_{2,n}^1$4 satisfying

$\mathcal{C}_{2,n}^1$5

The framework is $\mathcal{C}_{2,n}^1$6-rigid when only the trivial motions occur, and minimally $\mathcal{C}_{2,n}^1$7-rigid when it is both rigid and independent (Clinch et al., 2019).

2. Abstract rigidity and maximality in dimension three

Graver’s notion of an abstract $\mathcal{C}_{2,n}^1$8-rigidity matroid formalizes the closure properties shared by generic rigidity matroids. Nguyen’s characterization states that a matroid $\mathcal{C}_{2,n}^1$9 on $3$0, with $3$1, is an abstract $3$2-rigidity matroid if and only if every copy of $3$3 is a circuit and

$3$4

Consequently, abstract $3$5-rigidity matroids are precisely the matroids on $3$6 in which every copy of $3$7 is a circuit and the total rank is $3$8. Whiteley proved that the generic $3$9-cofactor matroid $G=(V,E)$0 is an abstract $G=(V,E)$1-rigidity matroid; in particular, $G=(V,E)$2 is a $G=(V,E)$3-matroid of rank $G=(V,E)$4 (Clinch et al., 2019).

Whiteley’s maximality conjecture proposed that, for every $G=(V,E)$5, the generic cofactor matroid $G=(V,E)$6 is the unique maximal abstract $G=(V,E)$7-rigidity matroid with respect to the weak order on matroids. Clinch, Jackson, and Tanigawa verified the case $G=(V,E)$8: the generic $G=(V,E)$9-cofactor matroid $p:V\to\mathbb{R}^2$0 is the unique maximal abstract $p:V\to\mathbb{R}^2$1-rigidity matroid, and more strongly the unique maximal $p:V\to\mathbb{R}^2$2-matroid on $p:V\to\mathbb{R}^2$3 (Clinch et al., 2019).

The three-dimensional proof is closely tied to matroid-erection theory. In the companion work on combinatorial characterization, $p:V\to\mathbb{R}^2$4 is identified as the free elevation of a rank-$p:V\to\mathbb{R}^2$5 matroid $p:V\to\mathbb{R}^2$6 whose non-spanning circuits are exactly the edge sets of copies of $p:V\to\mathbb{R}^2$7. A key step in the maximality proof is the verification that the double $p:V\to\mathbb{R}^2$8-replacement operation preserves independence in the generic $p:V\to\mathbb{R}^2$9-cofactor matroid, completing an inductive construction of bases analogous to Henneberg-type constructions in rigidity theory (Clinch et al., 2019).

This resolves Whiteley’s maximality conjecture in dimension three, but it does not resolve Graver’s separate conjecture that the generic three-dimensional bar-and-joint rigidity matroid $p(v_i)=(x_i,y_i)$0 is isomorphic to Whiteley’s cofactor matroid. That comparison remains active in later work.

3. Rank, independence, and combinatorial characterization

For $p(v_i)=(x_i,y_i)$1, independence admits a purely combinatorial characterization. A proper $p(v_i)=(x_i,y_i)$2-sequence is a finite sequence $p(v_i)=(x_i,y_i)$3 of $p(v_i)=(x_i,y_i)$4-circuits such that $p(v_i)=(x_i,y_i)$5 for all $p(v_i)=(x_i,y_i)$6. The rank of $p(v_i)=(x_i,y_i)$7 is

$p(v_i)=(x_i,y_i)$8

This gives a graph-theoretic rank formula for the maximal abstract $p(v_i)=(x_i,y_i)$9-rigidity matroid and solves the cofactor analogue of the combinatorial characterization problem for generic $s\ge 0$0-dimensional bar-joint rigidity (Clinch et al., 2019).

A second, cover-theoretic description uses $s\ge 0$1-thin, $s\ge 0$2-shellable covers. For a family $s\ge 0$3 of vertex sets, with hinges $s\ge 0$4, define

$s\ge 0$5

Then for $s\ge 0$6,

$s\ge 0$7

where the minimum is taken over all $s\ge 0$8 and all $s\ge 0$9-shellable, $\mathcal{C}_{d-1,n}^{d-2}$00-thin covers $\mathcal{C}_{d-1,n}^{d-2}$01 of $\mathcal{C}_{d-1,n}^{d-2}$02 with sets of size at least $\mathcal{C}_{d-1,n}^{d-2}$03. For a flat $\mathcal{C}_{d-1,n}^{d-2}$04, if $\mathcal{C}_{d-1,n}^{d-2}$05 denotes the maximal cliques of $\mathcal{C}_{d-1,n}^{d-2}$06 of size at least $\mathcal{C}_{d-1,n}^{d-2}$07, and $\mathcal{C}_{d-1,n}^{d-2}$08 the uncovered edges, then

$\mathcal{C}_{d-1,n}^{d-2}$09

Thus the rank of a flat is determined by the maximal cliques in the induced subgraph.

Several structural consequences follow. Every cyclic flat of $\mathcal{C}_{d-1,n}^{d-2}$10 is the union of copies of $\mathcal{C}_{d-1,n}^{d-2}$11. If $\mathcal{C}_{d-1,n}^{d-2}$12 is $\mathcal{C}_{d-1,n}^{d-2}$13-connected and $\mathcal{C}_{d-1,n}^{d-2}$14 is a cyclic set in $\mathcal{C}_{d-1,n}^{d-2}$15, then $\mathcal{C}_{d-1,n}^{d-2}$16 is $\mathcal{C}_{d-1,n}^{d-2}$17-rigid. If $\mathcal{C}_{d-1,n}^{d-2}$18 is $\mathcal{C}_{d-1,n}^{d-2}$19-connected and $\mathcal{C}_{d-1,n}^{d-2}$20 with $\mathcal{C}_{d-1,n}^{d-2}$21, then $\mathcal{C}_{d-1,n}^{d-2}$22 is $\mathcal{C}_{d-1,n}^{d-2}$23-rigid. The same work also notes that the rank problem for $\mathcal{C}_{d-1,n}^{d-2}$24 lies in NP$\mathcal{C}_{d-1,n}^{d-2}$25co-NP (Clinch et al., 2019).

4. Coincidence with bar-and-joint rigidity and hyperconnectivity

Whiteley’s cofactor matroid is not isolated from other rigidity-type matroids. For a configuration $\mathcal{C}_{d-1,n}^{d-2}$26, the bar-and-joint rigidity matroid $\mathcal{C}_{d-1,n}^{d-2}$27 is the row matroid of the standard rigidity matrix $\mathcal{C}_{d-1,n}^{d-2}$28. Kalai’s hyperconnectivity matroid $\mathcal{C}_{d-1,n}^{d-2}$29 is the row matroid of the hyperconnectivity matrix $\mathcal{C}_{d-1,n}^{d-2}$30. The cofactor matroid $\mathcal{C}_{d-1,n}^{d-2}$31 is the row matroid of the cofactor matrix $\mathcal{C}_{d-1,n}^{d-2}$32. All three are matroids on the ground set $\mathcal{C}_{d-1,n}^{d-2}$33 (Ruiz et al., 2021).

For distinct real numbers $\mathcal{C}_{d-1,n}^{d-2}$34, these three constructions coincide in a canonical family of realizations. The $\mathcal{C}_{d-1,n}^{d-2}$35-cofactor matroid of the points $\mathcal{C}_{d-1,n}^{d-2}$36 on the standard parabola in $\mathcal{C}_{d-1,n}^{d-2}$37, the bar-and-joint rigidity matroid of the points $\mathcal{C}_{d-1,n}^{d-2}$38 on the $\mathcal{C}_{d-1,n}^{d-2}$39-dimensional moment curve, and the $\mathcal{C}_{d-1,n}^{d-2}$40-hyperconnectivity matroid of the vectors $\mathcal{C}_{d-1,n}^{d-2}$41 all coincide. Because bar-and-joint rigidity and cofactor rigidity are projectively invariant, any non-degenerate conic may replace the parabola.

The coincidence is organized by the polynomial rigidity matroid $\mathcal{C}_{d-1,n}^{d-2}$42. If $\mathcal{C}_{d-1,n}^{d-2}$43 is the space of univariate polynomials of degree $\mathcal{C}_{d-1,n}^{d-2}$44, and for each vertex $\mathcal{C}_{d-1,n}^{d-2}$45 one chooses a basis

$\mathcal{C}_{d-1,n}^{d-2}$46

then the polynomial rigidity matrix has row $\mathcal{C}_{d-1,n}^{d-2}$47 equal to $\mathcal{C}_{d-1,n}^{d-2}$48 in block $\mathcal{C}_{d-1,n}^{d-2}$49, $\mathcal{C}_{d-1,n}^{d-2}$50 in block $\mathcal{C}_{d-1,n}^{d-2}$51, and zeros elsewhere. Its row matroid is independent of the chosen bases. Appropriate choices recover the three rigidity matrices: the basis $\mathcal{C}_{d-1,n}^{d-2}$52 gives bar-and-joint rigidity on the moment curve, the basis $\mathcal{C}_{d-1,n}^{d-2}$53 gives cofactor rigidity on the parabola, and the monomial basis $\mathcal{C}_{d-1,n}^{d-2}$54 gives hyperconnectivity.

This coincidence has additional algebraic consequences. In even dimension $\mathcal{C}_{d-1,n}^{d-2}$55, the generic hyperconnectivity matroid $\mathcal{C}_{d-1,n}^{d-2}$56 coincides with the algebraic matroid of skew-symmetric $\mathcal{C}_{d-1,n}^{d-2}$57 matrices of rank at most $\mathcal{C}_{d-1,n}^{d-2}$58. For $\mathcal{C}_{d-1,n}^{d-2}$59, $\mathcal{C}_{d-1,n}^{d-2}$60 is also the algebraic matroid of the Grassmannian.

5. Highly connected graphs and reconstruction from the matroid

A later line of work studies the three-dimensional cofactor matroid on an arbitrary simple graph $\mathcal{C}_{d-1,n}^{d-2}$61, written $\mathcal{C}_{d-1,n}^{d-2}$62. Because the original matrix definition is technically involved for the combinatorial arguments at hand, the rank characterization by $\mathcal{C}_{d-1,n}^{d-2}$63-shellable $\mathcal{C}_{d-1,n}^{d-2}$64-thin covers is taken as a definition. For simple graphs with at least five vertices, applying the rank formula with $\mathcal{C}_{d-1,n}^{d-2}$65 and $\mathcal{C}_{d-1,n}^{d-2}$66 yields the general upper bound $\mathcal{C}_{d-1,n}^{d-2}$67; $\mathcal{C}_{d-1,n}^{d-2}$68 is called $\mathcal{C}_{d-1,n}^{d-2}$69-rigid when equality holds (Garamvölgyi et al., 2022).

The same paper derives strong connectivity consequences. For the $\mathcal{C}_{d-1,n}^{d-2}$70-fold union $\mathcal{C}_{d-1,n}^{d-2}$71, if $\mathcal{C}_{d-1,n}^{d-2}$72 is $\mathcal{C}_{d-1,n}^{d-2}$73-connected, then

$\mathcal{C}_{d-1,n}^{d-2}$74

If $\mathcal{C}_{d-1,n}^{d-2}$75 is $\mathcal{C}_{d-1,n}^{d-2}$76-connected with $\mathcal{C}_{d-1,n}^{d-2}$77, then $\mathcal{C}_{d-1,n}^{d-2}$78 is vertically $\mathcal{C}_{d-1,n}^{d-2}$79-connected. In particular, when $\mathcal{C}_{d-1,n}^{d-2}$80, every $\mathcal{C}_{d-1,n}^{d-2}$81-connected graph is $\mathcal{C}_{d-1,n}^{d-2}$82-rigid and $\mathcal{C}_{d-1,n}^{d-2}$83 is vertically $\mathcal{C}_{d-1,n}^{d-2}$84-connected.

These connectivity estimates lead to Whitney-type reconstruction theorems. If $\mathcal{C}_{d-1,n}^{d-2}$85 is $\mathcal{C}_{d-1,n}^{d-2}$86-connected, $\mathcal{C}_{d-1,n}^{d-2}$87 has no isolated vertices, and $\mathcal{C}_{d-1,n}^{d-2}$88 is a matroid isomorphism between $\mathcal{C}_{d-1,n}^{d-2}$89 and $\mathcal{C}_{d-1,n}^{d-2}$90, then $\mathcal{C}_{d-1,n}^{d-2}$91 is induced by a graph isomorphism $\mathcal{C}_{d-1,n}^{d-2}$92. For $\mathcal{C}_{d-1,n}^{d-2}$93, a $\mathcal{C}_{d-1,n}^{d-2}$94-connected graph is therefore uniquely determined by its cofactor matroid. There is also a Servatius-type formulation: if $\mathcal{C}_{d-1,n}^{d-2}$95 is vertically $\mathcal{C}_{d-1,n}^{d-2}$96-connected and $\mathcal{C}_{d-1,n}^{d-2}$97 has no isolated vertices, then any isomorphism $\mathcal{C}_{d-1,n}^{d-2}$98 is induced by a graph isomorphism.

The same framework yields packing consequences. Every $\mathcal{C}_{d-1,n}^{d-2}$99-connected graph contains $K_n$00 edge-disjoint spanning subgraphs of cofactor rank $K_n$01, and every $K_n$02-connected graph has a spanning tree $K_n$03 such that $K_n$04 is $K_n$05-connected. These results extend the role of the cofactor matroid from a local rigidity model to a tool for graph reconstruction and high-connectivity structure theory.

6. Flexible circuits, implied nonedges, and current obstructions

Recent work studies Whiteley’s cofactor matroid in dimension three under the notation $K_n$06, the unique maximal matroid on $K_n$07 in which every graph isomorphic to $K_n$08 is a circuit. In that formulation, Whiteley’s cofactor matroid is exactly $K_n$09, and the open comparison is Graver’s maximality conjecture that $K_n$10 (Cheng et al., 17 Aug 2025).

This work isolates two notions that are especially relevant to the circuit structure. A nucleation of a graph is a rigid subgraph on at least five vertices; a graph is nucleation-free if it has no nucleation. A nonedge $K_n$11 of $K_n$12 is implied if $K_n$13 contains a circuit that includes $K_n$14. The resulting focus is on independent, nucleation-free graphs with implied nonedges, which are flexible but already encode nontrivial closure phenomena.

One basic family is the ring of butterflies $K_n$15. Each hinge in $K_n$16 is an implied nonedge; for $K_n$17, $K_n$18 is nucleation-free; for $K_n$19, $K_n$20 is independent and has $K_n$21 independent flexes; and for $K_n$22, $K_n$23 is dependent. The same paper proves that several operations preserve independence and nucleation-freeness, including $K_n$24-sums for $K_n$25, Henneberg-I, Henneberg-II under stated hypotheses, and $K_n$26-vertex splits for $K_n$27. More elaborate split-and-glue constructions, such as safe starting graphs and double-butterfly starting graphs, produce further independent, nucleation-free graphs with implied nonedges.

These constructions can also be combined to produce dependent graphs and circuits. If $K_n$28 and $K_n$29 are edge-disjoint graphs sharing only the endpoints of a nonedge $K_n$30, and $K_n$31 is implied in both, then $K_n$32 is dependent and $K_n$33 is implied in the union. If $K_n$34 and $K_n$35 are circuits and are otherwise vertex-disjoint, then $K_n$36 is a circuit.

None of these families refutes the conjecture $K_n$37, because their properties hold in both matroids. Their significance is different: they show that nucleation-free graphs with implied nonedges are key obstacles to settling Graver’s maximality conjecture and to obtaining a polynomial-time characterization of independence in $K_n$38. A related open possibility, stated explicitly in that work, is an $K_n$39-implied nonedge not contained in any $K_n$40 in the closure $K_n$41; such an example would be sufficient to refute the maximality conjecture.