---
title: r-Local Complementation in Graph Optimization
url: https://www.emergentmind.com/topics/r-local-complementation
type: topic
---

# r-Local Complementation in Graph Optimization

Searching arXiv for papers on local complementation, r-local minimum degree, and related graph-state/network results.
arxiv_search query: "local complementation graph state vertex-minor local minimum degree arXiv"
r-Local complementation denotes the study of graph transformations obtained by applying a bounded number $r$ of local complementations, together with the induced optimization and complexity questions that arise from this restriction. In the standard operation, for a graph $G=(V,E)$ and a vertex $u\in V$, local complementation replaces the induced subgraph on the neighborhood of $u$ by its complement. In the literature summarized here, this bounded-step perspective appears explicitly through the quantity $\delta_{\mathrm{loc}}^r(G)$, the minimum degree attainable after exactly $r$ local complementations [1503.04702]. By contrast, work on graph states and quantum routing uses only the ordinary one-vertex operation $\tau_a$ and does not introduce an “r-local complementation” generalization in that setting [1805.04559]. Related but distinct is edge local complementation, or pivot, which acts on an edge rather than a vertex and is used in Tanner-graph decoding while preserving the represented code [0905.4164].

## 1. Formal operation and bounded-step variants

Let $G=(V,E)$ be a simple undirected graph. For a vertex $u\in V(G)$, the local complementation of $G$ at $u$, denoted $G * u$ in [1503.04702], is obtained by replacing the induced subgraph $G[N_G(u)]$ by its complement. Equivalently, for all distinct $v,w\neq u$,
$$
\{v,w\}\in E(G*u) \iff \{v,w\}\notin E(G)\ \text{and}\ v,w\in N_G(u).
$$
An equivalent Boolean form given in the same source is
$$
v\sim_{G*u} w \iff (v\sim_G w)\oplus(v\sim_G u\wedge w\sim_G u).
$$
This is the standard vertex local complementation operation [1503.04702].

In graph-state notation, the same transformation is written as $\tau_a(G)$ for a vertex $a\in V$, where all edges between neighbors of $a$ are toggled and all other edges remain unchanged. If $\Gamma_G$ is the adjacency matrix over $\mathrm{GF}(2)$ and $\Theta_a$ is the adjacency matrix of the complete graph on $N_a$, then
$$
\Gamma_{\tau_a(G)}=\Gamma_G+\Theta_a \pmod 2.
$$
This formulation is used in quantum-network and graph-state analysis [1805.04559].

The bounded-step generalization appears in the definition of the $r$-local minimum degree:
$$
\delta_{\mathrm{loc}}^r(G)=\min\{\delta(G*_{u_1}*_{u_2}\cdots *_{u_r}):(u_1,\dots,u_r)\in V^r\}.
$$
Its basic properties are explicit: $\delta_{\mathrm{loc}}^0(G)=\delta(G)$, the sequence $\delta_{\mathrm{loc}}^r(G)$ is nonincreasing in $r$, and $\delta_{\mathrm{loc}}^\infty(G)=\delta_{\mathrm{loc}}(G)$ [1503.04702]. In this sense, r-local complementation is not a new elementary graph operation but a bounded-horizon regime for iterating the ordinary one.

A terminological distinction is necessary. In the quantum-network paper, “No r-local generalization in this paper” is stated explicitly; all constructions rely solely on $\tau_a$, which flips edges inside the 1-hop neighborhood $N_a$ [1805.04559]. Thus, “r-local complementation” in the bounded-step sense belongs to the optimization framework of local minimum degree rather than to the quantum-routing formalism of that paper.

## 2. Local minimum degree and equivalent characterizations

The central invariant associated with iterated local complementation is the local minimum degree
$$
\delta_{\mathrm{loc}}(G)=\min\{\delta(H): H\ \text{reachable from}\ G\ \text{by a sequence of local complementations}\}.
$$
This quantity asks how far the minimum degree can be reduced within the local-complementation orbit of a graph [1503.04702].

A key characterization uses the odd neighborhood of a vertex set. For $D\subseteq V$,
$$
\mathrm{Odd}_G(D)=\bigoplus_{u\in D} N_G(u),
$$
that is, a vertex belongs to $\mathrm{Odd}_G(D)$ iff it has an odd number of neighbors in $D$. The paper states the equivalence
$$
\delta_{\mathrm{loc}}(G)+1=\min_{D\neq\emptyset\subseteq V}|D\cup \mathrm{Odd}_G(D)|
=\min\{|A|:A\subseteq V,\ \mathrm{cutrk}_G(A)<|A|\},
$$
where $\mathrm{cutrk}_G(A)$ is the rank over $\mathbb F_2$ of the adjacency submatrix between $A$ and $V\setminus A$ [1503.04702].

These equalities connect local complementation to two distinct combinatorial viewpoints. The first is parity-based, through odd neighborhoods. The second is rank-based, through cut-rank. This places local complementation in direct contact with isotropic systems, rank-width methods, and vertex-minor theory. A plausible implication is that bounded-step questions such as $\delta_{\mathrm{loc}}^r(G)$ interpolate between purely local graph editing and full orbit optimization: the reachable set after $r$ steps is constrained, but the objective remains the same minimum-degree minimization.

The same parity-centric structure also underlies complexity reductions. In particular, the parameterized complexity results for LOCAL-MIN-DEGREE and BIP-LOCAL-MIN-DEGREE are formulated via the EVENSET problem, which asks for a nonempty set $D$ with $|D|\le k$ and $\mathrm{Odd}_G(D)=\emptyset$ in a bipartite graph [1503.04702]. This linkage shows that parity constraints are not merely representational convenience; they are structurally decisive.

## 3. Extremal bounds for $\delta_{\mathrm{loc}}$ and heuristic bounds for $\delta_{\mathrm{loc}}^r$

The extremal theory in [1503.04702] establishes both lower and upper bounds for the unconstrained local minimum degree. For every $n$, there exists an $n$-vertex graph $G$ with
$$
\delta_{\mathrm{loc}}(G)\ge 0.189n,
$$
and there exists a bipartite $n$-vertex graph with
$$
\delta_{\mathrm{loc}}(G)\ge 0.110n.
$$
On the upper-bound side, for any graph of order $n$,
$$
\delta_{\mathrm{loc}}(G)\le \frac{3n}{8}+o(n),
$$
and for bipartite graphs,
$$
\delta_{\mathrm{loc}}(G)\le \frac{n}{4}+o(n).
$$
The same paper also proves
$$
\delta_{\mathrm{loc}}(G)\le \tfrac12\,\tau(G)-\Theta(\log \tau(G)),
$$
where $\tau(G)$ is the vertex-cover number [1503.04702].

These bounds delimit how much degree reduction local complementation can enforce in the worst case. They also underpin exact exponential algorithms, because the search for a minimizing set can be truncated to subsets below these upper bounds.

For the bounded-step quantity $\delta_{\mathrm{loc}}^r(G)$, [1503.04702] gives a brute-force viewpoint and a heuristic combinatorial upper bound. By depth-$r$ search, one can explore all sequences of $r$ local complementations in $O(n^r)$ branches, pruning whenever the current minimum degree already exceeds the best value found. The paper then states, as a representative formula, that for every integer $r\ge 1$ and $n$ sufficiently large, any $G$ of order $n$ satisfies
$$
\delta_{\mathrm{loc}}^r(G)\le \frac{n}{2^{r+1}}+O(r\log n).
$$
This is presented as the outcome of iterating the proof strategy behind the general upper bound [1503.04702].

The status of this formula is important. It is given as a heuristic consequence of iterating the underlying argument rather than as a separately isolated theorem in the paper. This suggests a bounded-step decay picture in which even a small number of local complementations can substantially lower the minimum degree. A plausible implication is that many practical instances may be effectively reducible without traversing the full local-complementation orbit.

## 4. Complexity landscape and exact algorithms

The decision problem LOCAL-MIN-DEGREE asks, given a graph $G$ and integer $k$, whether $\delta_{\mathrm{loc}}(G)\le k$; BIP-LOCAL-MIN-DEGREE is the bipartite restriction. The complexity picture in [1503.04702] is layered.

First, the paper states that the local minimum degree problem is NP-Complete and hard to approximate. Second, it proves that EVENSET is FPT-reducible to LOCAL-MIN-DEGREE and that BIP-LOCAL-MIN-DEGREE is FPT-reducible to EVENSET. The resulting corollary is that LOCAL-MIN-DEGREE and BIP-LOCAL-MIN-DEGREE are FPT-equivalent to EVENSET. Consequently, they lie in W[2], while W[1]-hardness remains open because EVENSET’s hardness is itself open [1503.04702].

The exact-algorithm results are correspondingly exponential but nontrivial. For general graphs, $\delta_{\mathrm{loc}}(G)$ can be computed in time
$$
O^*(1.9387^n).
$$
For bipartite graphs, the sharper structural bound yields
$$
O^*(1.4667^n).
$$
These procedures enumerate subsets up to the proven size thresholds and test the cut-rank condition in polynomial time [1503.04702].

For bounded $r$, the paper provides a direct search procedure for $\delta_{\mathrm{loc}}^r(G)$ via depth-$r$ exploration of sequences of local complementations. The stated implementation idea is elementary but useful: recursively apply local complementation at every vertex, maintain the best minimum degree encountered at depth $r$, and prune whenever a branch cannot improve the current best [1503.04702]. Memoization of isomorphic states and pruning through global upper bounds are cited as natural refinements.

A separate but related complexity result arises in graph-state extraction. The problem “Given two labeled graphs $G,H$, can $|H\rangle$ be obtained from $|G\rangle$ by a sequence of local complementations and single-qubit Pauli measurements?” is equivalent to the VERTEX-MINOR problem and is NP-complete [1805.04559]. However, the same source identifies tractable cases: any three-party GHZ state can be distilled in polynomial time; certain four-party GHZ states can be distilled under a repeater-line condition; and if the source graph has bounded rank-width, then for any fixed target graph $H$ there is a polynomial-time algorithm deciding extractability and outputting the required sequence [1805.04559]. This places local complementation at a crossroads between difficult global orbit questions and restricted families admitting efficient algorithms.

## 5. Graph states, local Clifford implementation, and quantum routing

In graph-state quantum information, local complementation has an operational meaning at the level of state vectors. If $|G\rangle$ is the graph state associated with $G$, then local complementation at a vertex $a$ is implemented by a local Clifford unitary $U_a^\tau$ such that
$$
|\tau_a(G)\rangle=U_a^\tau|G\rangle,
$$
with
$$
U_a^\tau=(iX_a)^{1/2}\,(-iZ_{N_a})^{1/2},
$$
where $Z_{N_a}$ denotes $\prod_{b\in N_a} Z_b$ [1805.04559].

This equivalence between a combinatorial graph rewrite and a simple local Clifford unitary is central to quantum-network routing. The paper “Quantum network routing and local complementation” uses sequences of local complementations together with single-qubit Pauli measurements to reduce measurement overhead relative to standard repeater-style approaches [1805.04559]. The key theorem states:

> We can create an EPR pair between two nodes $a$ and $b$ of an arbitrary graph state using the $X$-protocol with at most as many measurements as with the standard repeater protocol.

The corresponding lemma states that $X$-measurements along a shortest path between two nodes are equivalent to performing a series of local complementations on that path, followed by $Z$-measurements on the intermediate nodes [1805.04559]. This identifies measurement-based routing primitives with local-complementation dynamics on the underlying graph.

The paper is explicit that it does not define an r-local extension: all constructions use only the standard 1-hop neighborhood operation $\tau_a$ [1805.04559]. Nevertheless, the bounded-step viewpoint from [1503.04702] provides a useful lens for interpreting such protocols. This suggests that some quantum-routing procedures can be regarded as carefully chosen short sequences in the local-complementation orbit, optimized not for minimum degree but for extractability of target entanglement resources.

## 6. Canonical examples and related operations

Two examples in [1805.04559] illustrate how local complementation changes routing capabilities.

In the butterfly network, the target is to create EPR pairs on $\{1,6\}$ and $\{2,5\}$. Rather than building two separate repeater lines, the optimal protocol is:
1. Perform local complementations in order: LC$_1$, LC$_3$, LC$_4$.
2. Then $Z$-measure nodes $3$ and $4$.
3. The remaining graph is two disjoint edges $(1,6)$ and $(2,5)$.

The paper states that this uses only two single-qubit measurements and bypasses the central bottleneck [1805.04559]. It also proves two structural propositions: there is no 5-node graph state with a bottleneck for simultaneous communication between two pairs of nodes that can be solved using local Cliffords and a Pauli measurement of a single node; and there are only four 6-node graph states with such a bottleneck solvable by local Cliffords and Pauli measurements, all equivalent to the butterfly up to relabeling [1805.04559].

In the $3\times 3$ cluster example, the goal is an EPR pair between vertices $1$ and $9$. The shortest path is $(1,2,5,6,9)$. The standard repeater approach requires isolating the path by $Z$-measuring all neighbors of the path, at least four nodes, followed by $X$-measuring the three intermediate nodes $2,5,6$, for a total of at least seven measurements. By contrast, the $X$-protocol measures $2,5,6$ in turn along the path and then $Z$-measures the new neighbors of $1$ and $9$, only three further nodes, for a total of six measurements. Moreover, one retains a residual 4-qubit cluster on $\{3,4,7,8\}$ from which a second EPR can be extracted by a single measurement [1805.04559].

A related but different operation is edge local complementation (ELC), also called pivot, as used in Tanner graphs [0905.4164]. For an edge $\{u,v\}$ in a bipartite Tanner graph, ELC toggles all edges between $\mathcal N_u^v=\mathcal N(u)\setminus\{v\}$ and $\mathcal N_v^u=\mathcal N(v)\setminus\{u\}$ and then swaps the adjacency lists of $u$ and $v$. This preserves the code because it corresponds over $\mathbb F_2$ to row additions and a column permutation of the parity-check matrix [0905.4164]. Although vertex local complementation and edge local complementation are distinct operations, both are orbit-generating local graph transformations with strong algebraic invariance properties.

## 7. Applications, scope, and common sources of confusion

The main application domains represented in the cited works are quantum communication and coding theory. In quantum networking, local complementation is used to solve bottlenecked simultaneous communication problems and to reduce the number of required measurements for extracting entanglement resources [1805.04559]. In coding theory, edge local complementation is interleaved with sum-product decoding in the SPA-ELC algorithm, where random pivots generate structurally distinct Tanner graphs for the same code and improve performance relative to standard SPA in the reported small-blocklength experiments [0905.4164].

Several distinctions are essential for avoiding conceptual conflation.

| Concept | Operation | Source |
|---|---|---|
| Local complementation | Complement induced subgraph on a vertex neighborhood | [1503.04702], [1805.04559] |
| r-local complementation | Bounded number $r$ of repeated local complementations, expressed through $\delta_{\mathrm{loc}}^r(G)$ | [1503.04702] |
| Edge local complementation | Pivot on an edge, toggle cross-neighborhood edges and swap adjacency lists | [0905.4164] |

A common misunderstanding is to treat “r-local complementation” as a standard graph-state operation defined in quantum-network work. The available source material does not support that interpretation. The quantum-routing paper explicitly states that no r-local generalization is introduced there; only the ordinary $\tau_a$ operation is used [1805.04559]. Conversely, the bounded-step notation $\delta_{\mathrm{loc}}^r(G)$ in [1503.04702] is about optimization of reachable minimum degree, not about redefining the elementary move itself.

Another possible source of confusion is the relationship between local complementation and extractability questions. In graph states, local complementations combined with Pauli measurements characterize vertex-minor reachability, and the general problem is NP-complete [1805.04559]. In ordinary graph optimization, local complementation alone defines the orbit over which $\delta_{\mathrm{loc}}$ and $\delta_{\mathrm{loc}}^r$ are minimized [1503.04702]. These are adjacent but not identical problems.

Taken together, the cited works show that local complementation is simultaneously a graph-rewriting primitive, a local Clifford action on graph states, and the basis for nontrivial optimization and complexity theory. The bounded-step perspective encapsulated by $\delta_{\mathrm{loc}}^r(G)$ formalizes how much can be achieved with only a limited number of such moves [1503.04702], while applications in quantum routing and code-preserving Tanner-graph transformations demonstrate that even short sequences of local operations can yield structurally and operationally significant effects [1805.04559] [0905.4164].

Source: https://www.emergentmind.com/topics/r-local-complementation