Geometric Rigidity Index in Planar Graphs

• Geometric Rigidity Index is a graph invariant defined via the normalized maximum independent edge set from the planar rigidity matroid, measuring proximity to generic rigidity in R².
• It abstracts geometric constraints into a combinatorial form, allowing efficient O(n²) computation and aiding wireless sensor network localization.
• The index distinguishes between achieving full rigidity rank and possessing redundant rigidity, clarifying conditions for global rigidity and unique network localization.

Searching arXiv for recent and directly relevant sources on “geometric rigidity index” and closely related rigidity notions. First, locating the primary graph-theoretic source that explicitly defines the rigidity index for wireless sensor network localization. Now checking broader arXiv uses of “geometric rigidity” to disambiguate the term across combinatorial rigidity, elasticity, and shell/thin-domain analysis. In graph rigidity and wireless sensor network localization, the geometric rigidity index most commonly denotes the rigidity index $K_r(G)$: a graph invariant that measures how close a graph is to being generically rigid in the plane. It is defined combinatorially from the planar rigidity matroid, not from a specific embedding, and therefore depends on graph structure under the usual genericity assumptions of $2$-dimensional rigidity theory. In broader rigidity literature, closely related language is also used for different objects—most notably thickness-scaling exponents in shell and thin-domain elasticity, and curvature-controlled quantities in variable-domain geometric rigidity—so the term is context-dependent rather than universal (Eren, 2015).

1. Definition in planar rigidity theory

Let $G=(V,E)$ be a graph. In the graph-localization setting, the key combinatorial notion is independence in the $2$-dimensional rigidity matroid: an edge set $E' \subseteq E$ is independent if, writing $V'$ for the set of vertices incident with $E'$, it satisfies

$$|E'| \le 2|V'| - 3 \text{ for all } E' \subseteq E.$$

If $S$ denotes the family of independent edge sets, then the rigidity index is

$K_r(G) := \frac{\max_{E' \in S} |E'|}{2|V|-3}.$

The denominator $2$0 is the maximum rank of the planar rigidity matroid, namely the number of edges in a minimally rigid graph in the plane (Eren, 2015).

The index takes values in $2$1. The interpretation is explicit: $2$2 for the empty graph, $2$3 for a graph that is not yet generically rigid but has accumulated some independent rigidity constraints, and $2$4 precisely when the graph attains full planar rigidity rank. In this sense, the quantity is a normalized rank and therefore a normalized measure of closeness to generic rigidity.

This definition is specific to generic rigidity in $2$5. It does not require node coordinates, and it is not an embedding-dependent measure of the rigidity of one particular realization. The quantity abstracts geometry into combinatorics.

2. Rigidity-theoretic and matroidal foundations

The rigidity index is grounded in the distinction between a framework and a graph. A framework $2$6 is a graph together with a placement $2$7 of the vertices in $2$8. A framework is rigid if it admits only trivial continuous deformations preserving edge lengths. A graph is rigid in $2$9 if every generic realization is rigid. Genericity is essential because it makes rigidity combinatorial rather than realization-specific (Eren, 2015).

At the infinitesimal level, the rigidity matrix $G=(V,E)$0 controls first-order deformations. For generic configurations in $G=(V,E)$1D, rigidity and infinitesimal rigidity coincide, and infinitesimal rigidity is characterized by

$G=(V,E)$2

Laman’s theorem gives the corresponding combinatorial criterion for minimal rigidity: $G=(V,E)$3 and

$G=(V,E)$4

In matroid language, independent sets satisfy the sparsity condition, while bases are maximal independent sets of size $G=(V,E)$5 (Eren, 2015).

A common misconception is to treat $G=(V,E)$6 as an edge-density statistic. It is not. The numerator counts the size of a maximum independent edge set, not the total number of edges. Extra edges beyond matroid rank do not increase $G=(V,E)$7. Accordingly, $G=(V,E)$8 certifies attainment of rigidity rank, but it does not distinguish a minimally rigid graph from an overconstrained rigid graph.

3. Computation and illustrative examples

Because the index is combinatorial, its input is only the graph structure: the vertex set $G=(V,E)$9 and edge set $2$0. Computation proceeds by finding a maximum-cardinality edge subset satisfying the planar sparsity inequalities and then normalizing by $2$1. The paper states that rigidity testing for an $2$2-vertex graph can be done in polynomial time, specifically $2$3, and that computing the rigidity index via independence testing also takes total running time $2$4 (Eren, 2015).

The standard examples make clear what the index does and does not record.

Graph	Rigidity status	$2$5
$2$6	Minimally rigid	$2$7
$2$8	Rigid, not minimal	$2$9
$E' \subseteq E$0	Non-rigid, all edges independent	$E' \subseteq E$1
$E' \subseteq E$2	Non-rigid, some generalized redundant edges	$E' \subseteq E$3

For $E' \subseteq E$4, the graph has exactly $E' \subseteq E$5 edges and all are independent, so $E' \subseteq E$6. For $E' \subseteq E$7, the graph is rigid but not minimally rigid; nevertheless $E' \subseteq E$8 because the rigidity rank is already saturated. This is the clearest example that the rigidity index is a rank measure rather than a minimality measure.

For the non-rigid graphs, the values are more revealing. In $E' \subseteq E$9, $V'$0, so $V'$1, and the maximum independent edge count is $V'$2, giving

$V'$3

In $V'$4, the number of independent edges is $V'$5, so

$V'$6

Both are non-rigid, but $V'$7 is combinatorially closer to rigidity.

4. Redundancy, global rigidity, and unique localizability

The rigidity index is only one part of the rigidity-theoretic picture. The same work introduces a broader notion of redundancy: an edge $V'$8 is a generalized redundant edge if

$V'$9

This definition applies even to non-rigid graphs. The associated redundancy index measures the fraction of such edges and complements $E'$0: the rigidity index measures closeness to rigidity rank, while the redundancy index measures how much of the edge set does not contribute to that rank (Eren, 2015).

This distinction matters because redundant rigidity is required for global rigidity, and in $E'$1D the paper recalls that a graph has a unique generic realization if and only if it is 3-connected and redundantly rigid. Consequently, $E'$2 implies generic rigidity, but it does not imply global rigidity, unique realization, or robustness to edge loss.

This point is especially important in localization. In wireless sensor networks, generic rigidity is associated with localizability, while redundant rigidity is associated with stronger uniqueness and robustness. The rigidity index is therefore best interpreted as a scalar summary of localization readiness, not as a complete certificate of unique localizability. A graph may attain full rigidity rank without being globally rigid, and it may be non-rigid while still containing generalized redundant edges, as $E'$3 shows.

5. Sensor-network design and sensing-radius transitions

The main application in the original graph-theoretic setting is cooperative localization in wireless sensor networks. The communication or measurement graph is modeled as a random geometric graph: $E'$4 nodes are placed uniformly at random, and an edge is added when two nodes lie within sensing radius $E'$5. Varying $E'$6 changes the edge set and therefore changes $E'$7 and the redundancy index $E'$8 (Eren, 2015).

Two simulation findings are particularly useful. In a sample graph with $E'$9 nodes in a $$|E'| \le 2|V'| - 3 \text{ for all } E' \subseteq E.$$0 area and $$|E'| \le 2|V'| - 3 \text{ for all } E' \subseteq E.$$1 ($$|E'| \le 2|V'| - 3 \text{ for all } E' \subseteq E.$$2), the reported values are

$$|E'| \le 2|V'| - 3 \text{ for all } E' \subseteq E.$$3

More generally, averaging over $$|E'| \le 2|V'| - 3 \text{ for all } E' \subseteq E.$$4 random geometric graph instances, the paper finds that $$|E'| \le 2|V'| - 3 \text{ for all } E' \subseteq E.$$5 reaches $$|E'| \le 2|V'| - 3 \text{ for all } E' \subseteq E.$$6 at about

$$|E'| \le 2|V'| - 3 \text{ for all } E' \subseteq E.$$7

whereas $$|E'| \le 2|V'| - 3 \text{ for all } E' \subseteq E.$$8 reaches $$|E'| \le 2|V'| - 3 \text{ for all } E' \subseteq E.$$9 at about

$S$0

The reported conclusion is that once rigidity is achieved, only about a 16% increase in sensing radius is needed on average to reach redundant rigidity.

The monotonicity behavior also differs. The paper notes that $S$1 is non-decreasing as $S$2 increases, whereas $S$3 is non-monotonic. This is consistent with their definitions: adding edges cannot reduce rigidity rank, but the fraction of edges that are generalized redundant can fluctuate as the graph densifies.

6. Broader uses of the term across rigidity theory

Outside combinatorial graph rigidity, the phrase “geometric rigidity” appears in several distinct senses. In thin-domain elasticity, Harutyunyan interprets the exponent $S$4 in a thickness-dependent best constant

$S$5

as the geometric rigidity of the domain, and derives an asymptotically sharp interpolation inequality with a singular coupling term of order $S$6 for thin domains built around a surface (Harutyunyan, 2019). In shell theory, Gaussian curvature plays the same classificatory role: for shells with positive Gaussian curvature the optimal linear rigidity constant scales like $S$7, while for negative Gaussian curvature it scales like $S$8; the scaling exponent thereby functions as an effective rigidity index of the shell class (Harutyunyan, 2016).

A different domain-dependent use appears for variable domains with free boundaries, where the controlling geometric quantity is a curvature-regularized surface energy

$S$9

and especially its curvature term $K_r(G) := \frac{\max_{E' \in S} |E'|}{2|V|-3}.$0, which determines the scale entering the rigidity estimate (Friedrich et al., 2021). In disordered central-force networks, no scalar geometric rigidity index is formally introduced, but the onset of rigidity is tracked by the smallest singular value of the equilibrium or compatibility matrix, which continuously vanishes when a state of self stress appears (Vermeulen et al., 2017).

This suggests that “geometric rigidity index” is not a single cross-disciplinary invariant. In graph localization it denotes a normalized planar rigidity-matroid rank; in shell and thin-domain elasticity it is naturally associated with sharp thickness-scaling exponents or curvature-controlled constants; and in geometry-sensitive network mechanics it may refer to singular-value diagnostics. The common theme is quantitative measurement of proximity to a rigid regime, but the mathematical object carrying that information depends strongly on the ambient theory.