Component Order Connectivity (COC)

• Component Order Connectivity (COC) is a graph vulnerability measure that evaluates how few deletions can fragment a graph into small connected components.
• It extends classical problems like Vertex Cover by incorporating weighted, directed, and exact-size variants, and uses kernelization and fixed-parameter tractability techniques.
• COC plays a key role in network reliability, spectral graph theory, and algorithmic graph decomposition, offering actionable insights for designing robust interconnection networks.

Component Order Connectivity (COC) is a graph-theoretic vulnerability measure centered on component order after deletions. In the standard unweighted formulation, given a graph $G$ and integers $k$ and $\ell$, the question is whether there exists a vertex set $S \subseteq V(G)$ with $|S| \le k$ such that the size of the largest connected component in $G-S$ is at most $\ell$ [(Kumar et al., 2016); (Drange et al., 2014)]. Closely related literature uses equivalent parameter language such as $k$-component order connectivity $\kappa_k(G)$, where one asks for the minimum number of vertices whose removal results in an induced subgraph in which every component has order at most $k-1$, and $k$0-component order edge connectivity $k$1, where edge deletions play the analogous role (Yatauro, 2023). Weighted, directed, exact-size, and lower-bounded-size variants extend the same basic idea by replacing vertex cardinality with weights, connected components with strongly connected components, the upper bound “at most $k$2” with the exact requirement “exactly $k$3,” or by requiring each post-deletion component to have order at least a threshold [(Drange et al., 2014); (Bang-Jensen et al., 2020); (Liu et al., 19 May 2026); (Zhang et al., 26 Sep 2025)]. This suggests that COC functions both as a specific deletion problem and as a broader component-order-aware connectivity paradigm.

1. Definitions and terminological scope

The standard unweighted decision problem asks whether, for a graph $k$4 and integers $k$5, there exists $k$6 with $k$7 such that every connected component of $k$8 has size at most $k$9 [(Drange et al., 2014); (Kumar et al., 2016)]. Equivalently, $\ell$0 is a hitting set for all connected vertex sets of size $\ell$1 (Kumar et al., 2016). In optimization form, one minimizes $\ell$2 subject to the maximum component order of $\ell$3 being at most $\ell$4 (Drange et al., 2014).

A closely aligned notation defines the $\ell$5-component order connectivity $\ell$6 as the minimum number of vertices whose removal results in an induced subgraph in which every component has order at most $\ell$7, and the $\ell$8-component order edge connectivity $\ell$9 as the minimum number of edges whose removal results in a subgraph in which every component has order at most $S \subseteq V(G)$0 (Yatauro, 2023). The shift by one is purely notational: the parameter $S \subseteq V(G)$1 in $S \subseteq V(G)$2 or $S \subseteq V(G)$3 corresponds to a maximum allowed component order of $S \subseteq V(G)$4.

Several canonical special cases anchor the subject. When $S \subseteq V(G)$5, the condition that every component in $S \subseteq V(G)$6 has exactly one vertex means that $S \subseteq V(G)$7 is an independent set, so COC reduces to Vertex Cover (Jain et al., 2024, Liu et al., 19 May 2026). When $S \subseteq V(G)$8 in the notation used for weighted and unweighted COC kernelization, the problem is again Vertex Cover (Casel et al., 2024). This identification is structurally important: many kernelization techniques for COC are explicit generalizations of crown-based reductions for Vertex Cover (Kumar et al., 2016, Liu et al., 19 May 2026).

The weighted formulation replaces cardinalities by vertex weights. If $S \subseteq V(G)$9 and $|S| \le k$0, then the weight of a heaviest component of a graph $|S| \le k$1 is

$|S| \le k$2

where $|S| \le k$3 are the connected components of $|S| \le k$4 (Drange et al., 2014). Weighted Component Order Connectivity (wCOC) asks whether there exists $|S| \le k$5 such that $|S| \le k$6 and $|S| \le k$7 (Drange et al., 2014). The paper introducing wCOC explicitly describes it as a refined version of weighted vertex integrity (Drange et al., 2014).

2. Principal variants and related parameters

The literature does not restrict COC to a single formulation. Instead, it contains a family of component-order-aware deletion parameters that differ in what is constrained after deletion: the largest component, every component, the exact order of every component, the number of components, or the minimum order of each component.

Variant	Post-deletion condition	Representative source
COC / $|S| \le k$8-COC	every connected component has size at most $|S| \le k$9	(Drange et al., 2014, Kumar et al., 2016)
wCOC	$G-S$0 and $G-S$1	(Drange et al., 2014)
Component order edge connectivity	every component has order at most $G-S$2 after edge deletions	(Yatauro, 2023)
Directed COC (DCOC)	every strongly connected component of $G-S$3 has size at most $G-S$4	(Bang-Jensen et al., 2020)
$G-S$5-Exact COC	every connected component in $G-S$6 has exactly $G-S$7 vertices	(Liu et al., 19 May 2026)
$G-S$8-extra $G-S$9-component connectivity	$\ell$0 has at least $\ell$1 components, each of order at least $\ell$2	(Zhang et al., 26 Sep 2025)

The exact variant sharpens the usual upper-bound requirement. In $\ell$3-Exact Component Order Connectivity, the question is whether one can delete at most $\ell$4 vertices so that every connected component in $\ell$5 has exactly $\ell$6 vertices (Liu et al., 19 May 2026). This specializes to Vertex Cover when $\ell$7, and to Deletion to Induced Matching when $\ell$8 (Liu et al., 19 May 2026).

A different branch of the literature constrains component order from below rather than above. The $\ell$9-extra $k$0-component connectivity $k$1 is the minimum number of vertices whose removal produces a disconnected graph with at least $k$2 components, where each component contains at least $k$3 vertices (Zhang et al., 26 Sep 2025). Likewise, $k$4-good $k$5-component connectivity $k$6 requires at least $k$7 components and, in addition, every remaining vertex to have at least $k$8 neighbors in the remaining graph; this implies that every component of $k$9 has order at least $\kappa_k(G)$0 (Ding et al., 2024). These lower-bounded-size formulations are not the same as standard COC, but they are explicitly component-order-aware.

Component connectivity, by contrast, constrains the number of components rather than their order. For a non-complete connected graph $\kappa_k(G)$1, a $\kappa_k(G)$2-component cut is a set $\kappa_k(G)$3 such that $\kappa_k(G)$4 has at least $\kappa_k(G)$5 connected components, and

$\kappa_k(G)$6

(Zhao et al., 2018). The folded-hypercube paper explicitly states that in COC language, the basic COC without order constraints is exactly this notion (Zhao et al., 2018). This places component connectivity and COC in the same reliability family, but with different emphasis: number of components versus maximum or minimum component order.

3. Algorithmic complexity and kernelization

The algorithmic landscape of COC is sharply stratified by graph class, weighting, and parameterization. On arbitrary graphs, wCOC is weakly NP-complete already on complete graphs, COC is NP-complete on split graphs, and COC is W[1]-hard on split graphs when parameterized by $\kappa_k(G)$7 or by $\kappa_k(G)$8 (Drange et al., 2014). For the combined parameter $\kappa_k(G)$9, however, wCOC is fixed-parameter tractable: it can be solved in

$k-1$0

time, and it admits a kernel with at most

$k-1$1

vertices; moreover, there is no algorithm of time $k-1$2 for COC unless the Exponential Time Hypothesis fails (Drange et al., 2014).

On interval graphs the situation is more favorable. Weighted COC can be solved in

$k-1$3

time, while the unweighted version can be solved in $k-1$4 time on this class (Drange et al., 2014). These algorithms use interval models, clique paths, and dynamic programming over separator structure (Drange et al., 2014).

Kernelization has been a central theme. A linear-programming-based kernel with at most $k-1$5 vertices was proved for $k-1$6-COC, initially in $k-1$7 time for every constant $k-1$8, and then with a separation oracle implying running time $k-1$9 (Kumar et al., 2016). Later work on vulnerability measures improved the weighted COC kernel from $k$00 to

$k$01

and also gave a combinatorial algorithm that provides a $k$02 vertex kernel in FPT-runtime when parameterized by $k$03, where $k$04 is the size of a maximum $k$05-packing (Casel et al., 2024). The same work shows that this $k$06 kernelization can be transformed into a polynomial algorithm for the special cases $k$07 and claw-free graphs (Casel et al., 2024).

The exact-size variant also admits linear kernels. $k$08-Exact Component Order Connectivity has a kernel with

$k$09

vertices computable in $k$10 time (Liu et al., 19 May 2026). This yields a $k$11-vertex kernel for Vertex Cover when $k$12, and a $k$13-vertex kernel for Deletion to Induced Matching when $k$14, improving the previously known $k$15-vertex kernel for that case (Liu et al., 19 May 2026).

These kernelizations rely on generalized crown structures. The 2016 $k$16 kernel uses a linear programming relaxation together with a weighted generalization of the $k$17-Expansion Lemma (Kumar et al., 2016). The 2024 work combines balanced crown decomposition, demanded balanced expansions, and weighted crown reductions to unify and extend earlier approaches for VI, wVI, and wCOC (Casel et al., 2024). The 2026 exact-size kernel introduces ECOC crown decompositions in which the crown side consists of connected components of size exactly $k$18 rather than an independent set (Liu et al., 19 May 2026).

4. Degree conditions, extremal structure, and spectral viewpoints

COC also has a strong extremal side. For the vertex version, degree sequences can force lower bounds on $k$19-component order connectivity. If $k$20 and $k$21, then the degree-sequence condition

$k$22

implies that every realization is forcibly $k$23-component order $k$24-connected (Yatauro, 2023). For the edge version, if $k$25 and $k$26, then

$k$27

implies that every realization is forcibly $k$28-component order $k$29-edge connected (Yatauro, 2023). These statements are part of a broader program of best monotone theorems based on degree-sequence majorization and sink characterization (Yatauro, 2023).

The same paper shows that the complexity of optimal degree conditions grows rapidly. For $k$30 and $k$31, there are at least $k$32 sinks for “$k$33-component order $k$34-edge connected,” where $k$35 is the integer partition function (Yatauro, 2023). This links optimal monotone criteria for COC-style edge connectivity to the combinatorics of integer partitions (Yatauro, 2023).

Spectral extremal graph theory has recently incorporated component-order-aware connectivity constraints. For graphs of order $k$36, minimum degree $k$37, and fixed $k$38-good $k$39-component connectivity $k$40, the maximum adjacency spectral radius is attained by explicit join-of-cliques constructions, with six cases depending on inequalities among $k$41 (Ding et al., 2024). The extremal graphs are described in the paper as having the general form of a “small clique representing the cut” joined to “one big clique” and $k$42 equal cliques (Ding et al., 2024). For $k$43-extra $k$44-component connectivity $k$45, the corresponding extremal problem for the distance spectral radius likewise yields join constructions, with distinct optimal forms in the regimes $k$46, $k$47, and $k$48 (Zhang et al., 26 Sep 2025). These results show that component-order constraints are compatible with sharp spectral extremal characterizations.

5. Reliability, interconnection networks, and other applications

A principal motivation for component-order-aware connectivity is network reliability. In interconnection networks, vertices model processors and edges model communication links; classical connectivity only detects the first disconnection threshold, whereas component-based measures quantify how rapidly the network fragments into many or small pieces under faults (Zhao et al., 2018). The folded-hypercube study explicitly argues that traditional connectivity “always underestimates the resilience of large networks,” while component connectivity “can more accurately evaluate the reliability and fault tolerance for large-scale parallel processing systems” (Zhao et al., 2018).

For the $k$49-dimensional folded hypercube $k$50, the paper determines the $k$51-component connectivity

$k$52

for $k$53 and $k$54 (Zhao et al., 2018). This formula generalizes ordinary connectivity, since $k$55 (Zhao et al., 2018). The same work also states that the formula fails at $k$56, and that determining $k$57 for $k$58 remains open (Zhao et al., 2018).

Alternating group graphs and split-stars provide additional exact component-fragmentation thresholds. For $k$59,

$k$60

and for $k$61,

$k$62

(Gu et al., 2018). These formulas belong to component connectivity rather than upper-bounded-order COC, but they serve the same reliability objective of quantifying resistance to fragmentation.

COC also appears outside reliability theory. In the cops and robber game, the paper on 2-component order connectivity defines $k$63-coc$k$64 as the size of a smallest set $k$65 such that all the connected components of the induced graph on $k$66 are of size at most $k$67 (Jain et al., 2024). Specializing to $k$68, it proves the bound

$k$69

for the cop number $k$70 (Jain et al., 2024). This result turns a component-order deletion parameter into a pursuit-evasion bound.

6. Directed, exact, and open directions

The directed analogue replaces connected components by strongly connected components. Directed Component Order Connectivity (DCOC) asks, for a digraph $k$71 and integers $k$72, whether there exists $k$73 of size $k$74 such that the largest strongly connected component in $k$75 has at most $k$76 vertices (Bang-Jensen et al., 2020). For $k$77, DCOC reduces to Directed Feedback Vertex Set (Bang-Jensen et al., 2020). On general digraphs, parameter $k$78 admits an $k$79-time algorithm, improving the previous $k$80 bound (Bang-Jensen et al., 2020). On semicomplete digraphs, DCOC parameterized by $k$81 can be solved in time $k$82, and there is no algorithm of time $k$83 unless the Exponential Time Hypothesis fails; the same paper also gives matching lower bounds for parameter $k$84 (Bang-Jensen et al., 2020).

Exact-size constraints form another frontier. $k$85-Exact Component Order Connectivity asks whether every connected component in $k$86 can be forced to have exactly $k$87 vertices (Liu et al., 19 May 2026). Its kernelization via ECOC crown decompositions generalizes both the classical Vertex Cover crown decomposition and the induced-matching case, and the paper identifies the running-time dependence on $k$88 as a limitation: obtaining an $k$89-vertex kernel using time also polynomial in $k$90 is left open (Liu et al., 19 May 2026). The same paper also identifies establishing non-trivial lower bounds on kernels for this exact variant as an open topic (Liu et al., 19 May 2026).

More broadly, current open directions arise at the interface between component order, component count, and application-specific constraints. For folded hypercubes, determining high-$k$91 component connectivity beyond the range $k$92 remains open (Zhao et al., 2018). For cops and robber, a natural next step is to bound the cop number by $k$93-coc for a constant $k$94, possibly with a coefficient better than $k$95 (Jain et al., 2024). The spectral literature suggests further extensions to signless Laplacian, Laplacian, directed analogues, and edge-based versions of component-order-aware connectivity, but the decisive structural fact already visible is that many extremal problems continue to produce join-of-cliques architectures under COC-type constraints (Ding et al., 2024, Zhang et al., 26 Sep 2025).

Taken together, these developments show that Component Order Connectivity is not a single invariant but a mature cluster of deletion parameters and decision problems. Its core question—how many failures are needed before all surviving components become small, or before multiple components of controlled order must appear—links parameterized complexity, crown-based kernelization, degree-sequence extremal theory, spectral graph theory, network reliability, and algorithmic graph decomposition.