2-Local Complementation in Graph Theory

• 2-local complementation is a graph operation that applies three successive local complementations (edge pivot) to rewire graphs while preserving equivalence classes.
• It enables efficient transformations in quantum graph states and error-correcting codes via local-Clifford operations and vertex-minor extraction.
• The technique underpins network routing protocols and orbit theory analyses, and relates to NP-complete minimization and P-complete evaluation problems.

Searching arXiv for the supplied topic and ids to ground the article in recent and relevant papers. 2-local complementation is a context-dependent term in graph theory and quantum information. In the most common usage it denotes edge local complementation—also called pivot—on an edge $uv$, given by the three-step composition

$$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$

where $\tau_u$ is local complementation at a single vertex $u$. In qutrit graph-state theory, however, the phrase can also refer to the $\lambda=2$ instance of weighted local complementation over $\mathbb{F}_3$. Across these usages, the unifying idea is that local graph rewiring generated by local complementations organizes local-Clifford equivalence, vertex-minor extraction, code equivalence, and several network-routing procedures (Joo et al., 2011, Hahn et al., 2018).

1. Scope, definitions, and terminological ambiguity

For a simple graph $G=(V,E)$ and a vertex $a\in V$, local complementation replaces the induced subgraph on the neighborhood $N_a$ by its complement. In the adjacency-matrix formulation used for graph states, if $\Gamma_G$ is the adjacency matrix and $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$0 is the complete graph on $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$1, then

$$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$2

Equivalently, every edge between distinct neighbors of $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$3 is toggled, while no other edge is changed (Hahn et al., 2018).

The most standard graph-theoretic meaning of 2-local complementation is the edge-centered operation obtained by composing three local complementations on two vertices. In the graph-state and coding literature this appears as

$$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$4

and is described as edge local complementation on the edge $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$5 (Joo et al., 2011). The same operation is also written as a pivot: $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$6 This is the sense in which 2-local complementation is usually tied to a pair of vertices or to an edge (0710.2243).

A persistent source of ambiguity is that the phrase is not uniform across subfields. Some graph- and matroid-oriented discussions describe ordinary local complementation itself as “2-local” because it toggles adjacency among pairs of neighbors of a distinguished vertex (Oxley et al., 2019). By contrast, the qutrit ZX-calculus and qutrit graph-state literature reserve “2-local complementation” for the weighted operation with parameter $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$7,

$$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$8

which is one of the two qutrit local-complementation variants $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$9 (Gong et al., 2017). A common misconception is therefore to treat “2-local complementation” as a single universally fixed primitive; the literature instead supports at least two established readings, both derived from the same local-complementation framework.

2. Algebraic and matroidal structure

Local complementation, pivot, and loop complementation admit a compact algebraic description over $\tau_u$0. For set systems and graphs with loops allowed, single-vertex pivot and loop complementation are represented by the matrices

$\tau_u$1

and the operations they generate on a fixed vertex form the permutation group $\tau_u$2. This yields normal forms for sequences of pivots and loop complementations and gives an alternative proof of the classic identity equating edge complementation with a triple local complementation on simple graphs (0909.4004).

From this viewpoint, edge local complementation is not an independent primitive but a short word in the local generators. On simple graphs,

$\tau_u$3

so the 2-vertex operation is generated by three 1-vertex operations (0909.4004). This identity underlies much of the later coding and graph-state literature: a pairwise rewiring can be studied either as a pivot or as a constrained path in an LC orbit.

The isotropic matroid $\tau_u$4 provides a classification framework for local equivalence. For a looped simple graph $\tau_u$5, one forms

$\tau_u$6

whose columns are partitioned into vertex triples $\tau_u$7. Traldi’s main result is that $\tau_u$8 classifies $\tau_u$9 up to local equivalence; in particular, for simple graphs,

$u$0

Within this formalism, loop complementation, non-simple local complementation, and edge pivots become local $u$1-permutations on the corresponding vertex triples (Traldi, 2013).

This algebraic rephrasing has two consequences. First, it shows that 2-local complementation in the edge-pivot sense is already built into the same local symmetry group as ordinary local complementation. Second, it furnishes invariants—matroid isomorphism type, Tutte-type specializations, and associated delta-matroids and isotropic systems—that are stable under these operations (Traldi, 2013).

3. Graph states and local Clifford implementations

For a graph $u$2, the associated qubit graph state is

$u$3

In this setting local complementation is exactly a local Clifford transformation. If $u$4, then

$u$5

Hence every LC move is physically a product of single-qubit Clifford gates, and LC-equivalence is the graph-theoretic expression of local-Clifford equivalence (Hahn et al., 2018).

The edge-pivot interpretation appears naturally when Pauli measurements are rewritten as graph transformations. In the routing analysis of graph states, an $u$6-measurement on a vertex $u$7 with chosen neighbor $u$8 is expressed as

$u$9

On the graph this is exactly the three-step pattern

$\lambda=2$0

followed by deletion of $\lambda=2$1 via $\lambda=2$2. In graph-theoretic language, this is a pivot on the edge $\lambda=2$3 plus deletion of one endpoint (Hahn et al., 2018). The same mechanism generalizes along a path: repeated $\lambda=2$4-measurements on intermediate vertices are equivalent to a chain of local complementations followed by $\lambda=2$5-measurements on the deleted vertices.

A related but more explicit physical realization is given for edge local complementation itself. If two graph states are connected by a single $\lambda=2$6 gate on core qubits $\lambda=2$7, and the neighborhoods of $\lambda=2$8 and $\lambda=2$9 are disjoint, then

$\mathbb{F}_3$0

In that regime, edge local complementation on the edge $\mathbb{F}_3$1 is implemented by two Hadamard gates on the endpoints (Joo et al., 2011). This identity is a particularly transparent operational meaning of 2-local complementation: a nontrivial graph rewrite concentrated on two vertices becomes an inexpensive local Clifford circuit.

4. Networks, codes, and state-preparation applications

In quantum-network routing, local complementation is used to reduce measurement overhead and to reconfigure entanglement before deletion of unwanted vertices. For a shortest path $\mathbb{F}_3$2, the $\mathbb{F}_3$3-protocol of graph-state routing first $\mathbb{F}_3$4-measures the intermediate vertices and then $\mathbb{F}_3$5-measures the remaining neighbors of the endpoints. The central theorem states that this protocol creates an EPR pair between arbitrary nodes $\mathbb{F}_3$6 and $\mathbb{F}_3$7 with at most as many measurements as the repeater protocol, and it is precisely the decomposition of pathwise $\mathbb{F}_3$8-measurements into local-complementation chains plus deletions that gives the advantage (Hahn et al., 2018).

The butterfly network is the canonical bottleneck example. There the desired simultaneous pairs $\mathbb{F}_3$9 and $G=(V,E)$0 are obtained by consecutive local complementations on nodes $G=(V,E)$1, $G=(V,E)$2, and $G=(V,E)$3, followed by $G=(V,E)$4-measurements on $G=(V,E)$5 and $G=(V,E)$6. In the 9-qubit cluster example, local complementations at nodes $G=(V,E)$7 followed by deletion of $G=(V,E)$8 realize the same effect as a pathwise $G=(V,E)$9-measurement protocol while preserving a residual entangled resource (Hahn et al., 2018). These examples are often best interpreted as chained 2-local moves assembled from edge-pivot patterns.

Binary linear coding theory provides another canonical application. If a binary $a\in V$0 code has generator matrix $a\in V$1, then it corresponds to the $a\in V$2-bipartite graph with adjacency matrix

$a\in V$3

The ELC orbit of this bipartite graph is exactly the equivalence class of the code under coordinate permutation (0710.2243). More specifically, a single ELC on an edge $a\in V$4, together with the endpoint swap built into the definition, corresponds to swapping two code coordinates and then restoring standard form by row operations.

This orbit viewpoint encodes standard coding data. Each labeled graph in the ELC orbit corresponds to an information set of the code, and if the code is self-dual then each labeled bipartite graph corresponds to two information sets. The minimum distance $a\in V$5 is recovered from the orbit by

$a\in V$6

where $a\in V$7 is the smallest vertex degree attained by a vertex in the partition of size $a\in V$8 over the whole ELC orbit (0710.2243). The class of ELC-preserved graphs consists of graphs with ELC orbit of size one; exhaustive search identifies all such graphs up to order $a\in V$9, and all bipartite ELC-preserved graphs up to order $N_a$0 (Danielsen et al., 2010).

The same 2-local machinery also yields efficient encoded-state constructions. For logical cluster states based on the five-qubit code, a single physical $N_a$1 together with Hadamards implementing edge local complementation suffices to realize a logical $N_a$2 between two logical qubits. The resulting construction reduces the number of required entangling gates relative to the naive implementation and extends to hierarchical encoded graph states (Joo et al., 2011).

5. Orbit theory and nonbinary extensions

LC and ELC are most naturally organized through orbit structure. For qubit graph states, repeated application of local complementation generates the full local-Clifford equivalence class, and orbit graphs up to $N_a$3 qubits reveal systematic correlations between orbit-theoretic observables and entanglement measures. In particular, orbit diameter, chromatic number, and related graph-theoretic quantities correlate with Schmidt measure, rank-width, and preparation complexity (Adcock et al., 2019). This suggests that the geometry of an LC orbit is itself a useful descriptor of entanglement class.

A more elaborate invariant is the graph $N_a$4 built from the stabilizer events of the graph state associated with $N_a$5: its vertices are the stabilizer-consistent events and its edges encode exclusivity. If $N_a$6 and $N_a$7 lie in the same Kotzig orbit, then

$N_a$8

and conversely the same $N_a$9 arises only from graphs in the same orbit (Cabello et al., 2012). For connected $\Gamma_G$0 with $\Gamma_G$1,

$\Gamma_G$2

linking LC orbits to contextuality, Bell inequalities, and entanglement-assisted zero-error capacity (Cabello et al., 2012).

The nonbinary case modifies both the definitions and the generators. For qutrit graph states, the adjacency matrix $\Gamma_G$3 is symmetric over $\Gamma_G$4, edge weights lie in $\Gamma_G$5, and local Clifford equivalence on graphs is generated not by local complementation alone but by local complementation together with local scaling. The qutrit local-complementation rule is

$\Gamma_G$6

and orbit classification up to $\Gamma_G$7 qutrits shows strong correlations between orbit-graph connectivity and Schmidt measure (Revis et al., 5 Jun 2025). In this framework, the term “2-local complementation” does not denote an edge pivot; all primitive moves remain single-vertex operations, and the paper explicitly states that it does not introduce a two-vertex complementation primitive (Revis et al., 5 Jun 2025).

A different nonbinary usage appears in the qutrit ZX-calculus. There, local complementation is parameterized by $\Gamma_G$8, so the $\Gamma_G$9 case is literally called 2-local complementation. The local-complementation property for qutrit graph states is equivalent to an Euler decomposition of the qutrit Hadamard gate into $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$00-green and red rotations, and the paper states that the $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$01 case is proved in the same way as the $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$02 case (Gong et al., 2017). This is a second, conceptually distinct, meaning of the term.

6. Complexity, bounds, and algorithmic status

The most developed complexity theory in this area concerns the local minimum degree

$$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$03

which measures the smallest minimum degree attainable within the LC orbit. It admits the characterization

$$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$04

where

$$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$05

This quantity is NP-complete to decide and admits no constant-factor approximation unless $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$06, even for bipartite graphs (Javelle et al., 2012).

The same line of work gives both lower and upper extremal bounds. For Paley graphs $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$07,

$$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$08

Probabilistic arguments yield infinitely many graphs with local minimum degree at least $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$09, and infinitely many bipartite graphs with local minimum degree at least $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$10 (Javelle et al., 2012). On the upper side, every graph of order $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$11 satisfies

$$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$12

and every bipartite graph satisfies

$$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$13

There is also the bound that $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$14 is smaller than half of the vertex cover number up to a logarithmic term (Cattanéo et al., 2015).

Parameterized and exact algorithms refine this picture. LOCAL MINIMUM DEGREE and its bipartite restriction are in $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$15 and are FPT-equivalent to EVENSET, whose $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$16-hardness remains a long-standing open question (Cattanéo et al., 2015). Exact algorithms compute $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$17 in time $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$18 for general graphs and $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$19 for bipartite graphs (Cattanéo et al., 2015).

A distinct but complementary complexity result concerns explicit evaluation of a prescribed LC sequence. Given a graph $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$20, a sequence of vertices $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$21, and a pair $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$22, the Local Complementation Problem asks whether $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$23 is present in $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$24. This problem is $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$25-complete; for complete and star graphs it drops to $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$26, and the paper conjectures $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$27-completeness also for circle graphs (Concha-Vega, 31 Mar 2025). Because edge local complementation is a fixed short composition of local complementations, this suggests that algorithmic questions about 2-local complementation in the edge-pivot sense inherit the same essentially sequential character.

Taken together, these results delimit the subject sharply. Local and 2-local complementations are algebraically compact and physically implementable, yet the orbit structure they generate is rich enough to support nontrivial classification theorems, coding correspondences, and routing protocols, while also exhibiting NP-complete minimization problems and $$G * uv \;=\; \tau_u \circ \tau_v \circ \tau_u(G) \;=\; \tau_v \circ \tau_u \circ \tau_v(G),$$28-complete forward-evaluation problems.