PTAS for Planar k-Connectivity Augmentation

Updated 31 December 2025

The paper introduces PTAS/EPTAS algorithms that near-optimally augment planar graphs for constant k using decomposition, bounded-treewidth, and dynamic programming techniques.
It tackles both vertex- and edge-connectivity variants by leveraging structural properties like separating triangles and laminar k-cuts to address NP-hard challenges.
The study extends its methods to triangulation models and beyond-planar classes, detailing algorithmic trade-offs and establishing clear (1+ε) approximation guarantees.

The planar $k$ -connectivity augmentation problem (PTAS for Planar $k$ -Connectivity Augmentation) concerns finding a minimum-size subset of edges whose addition to a given planar graph increases its connectivity to a prescribed $k$ . In both vertex and edge models, this augmentation problem is NP-hard for small $k$ in the strict planar setting, yet recent research has yielded PTAS (and EPTAS) frameworks for constant $k$ in planar graphs by leveraging structural properties, decomposition techniques, and bounded-treewidth algorithms. Connections to flip distance in triangulations and extensions to beyond-planarity further enrich this domain.

1. Formal Problem Definitions and Variants

Given a $c$ -vertex-connected or $k$ -edge-connected planar graph $G = (V, E)$ and a target connectivity $k > c$ , the planar $k$ -connectivity augmentation problem asks for a minimum-cardinality set $F$ of new edges such that the augmented graph $G' = (V, E \cup F)$ attains $k$ -connectivity and, in strict variants, remains planar (Akitaya et al., 1 Sep 2025, Neuwohner et al., 24 Dec 2025). For triangulations, the problem is equivalent to finding a minimum-length sequence of edge flips transforming $T$ into a $k$ -connected triangulation; the flip distance model directly maps to connectivity augmentation.

Two primary variants are addressed:

Variant	Augmentation Goal	Constraint
Vertex-connectivity (VCA)	$k$ -vertex-connected $G'$	Planarity, triangulation, PSLG
Edge-connectivity (ECA)	$k$ -edge-connected $G'$	Planarity, candidate links

For $k\leq 5$ , planarity imposes combinatorial restrictions, and NP-completeness holds for $2\leq c<k\leq 5$ (Akitaya et al., 1 Sep 2025, Neuwohner et al., 24 Dec 2025). When augmentation is permitted into $\ell$ -planar classes ( $\ell=\Theta(k^2)$ ), polynomial-time schemes become feasible.

2. Hardness Results and Structural Barriers

The $c \to k$ connectivity augmentation problem in planar graphs is NP-complete for $2 \leq c < k \leq 5$ (Akitaya et al., 1 Sep 2025), and more generally, planar $k$ -CAP is NP-hard for all $k \geq 2$ (Neuwohner et al., 24 Dec 2025). These results are established via reductions from Linked Planar 3-SAT, with gadget constructions scaling with $k$ .

Structural properties that underpin this hardness include the existence of separating triangles (cycles), long terminal-free cycles in minimal $k$ -vertex-connected planar graphs for $k\geq 4$ , and the laminarity of $k$ -cuts:

Obstacle	Manifestation
Separating triangles	Induce required edge/flip operations for augmentation
Terminal-free cycles	Break down of local-to-global patching in PTAS frameworks
Laminar $k$ -cuts	Necessitate careful handling of cuts across decomposition

This structural complexity explains why canonical PTAS decompositions (e.g., Baker’s) require new innovations for global connectivity objectives.

3. PTAS and EPTAS Frameworks

Recent advances have introduced PTAS and EPTAS architectures for planar $k$ -connectivity augmentation for constant $k$ , including specialized schemes for triangulations and general planar graphs (Akitaya et al., 1 Sep 2025, Borradaile et al., 2016, Neuwohner et al., 24 Dec 2025).

Triangulation EPTAS (Flip Distance)

For $c$ -connected triangulations, connectivity augmentation (to $k=4$ ) via edge flips is addressed by reducing the problem to hitting separating triangles (3-cycles). Any flip sequence must destroy each separating triangle, yielding the optimal hitting set $\tau$ :

Hitting set $\tau = \min |E'|$ intersecting all separating triangles.
EPTAS computes a sequence of at most $(1+\epsilon)\cdot\tau$ flips in $O(n \cdot 2^{O(1/\epsilon)})$ time (Akitaya et al., 1 Sep 2025).
Utilizes Baker-style layered decomposition: edges are colored/modulo layered, and low-impact subproblems are solved via DP on small treewidth pieces.

General $k$ -CAP PTAS

For general planar $k$ -edge-connectivity augmentation:

Dual-layer (“ring”) decomposition of the vertex–face graph, partitioning into overlapping low-treewidth sets $U_i$ with controlled edge overlap.
Definition of “ $k$ -edge-safe cover”: for every cut $S$ , either enough edges are bought to cross the cut, or the cut lies entirely inside one piece.
Chain graphs of snug vertices and laminar families of $k$ -cuts enable effective preprocessing and contraction steps.
Bounded-treewidth dynamic programming solves local pieces, with global feasibility achieved by gluing via overlapping edge purchases.
Approximation factor $(1+\epsilon)$ attained for constant $k$ in $O(n)$ time, with singly exponential dependence on $k/\epsilon$ (Neuwohner et al., 24 Dec 2025).

4. Key Structural Lemmas and Decomposition Principles

Successful application of PTAS/EPTAS techniques in connectivity augmentation hinges on careful exploitation of planar structure, minimal $k$ -connectivity, and bounded treewidth (Borradaile et al., 2016, Neuwohner et al., 24 Dec 2025):

Tree-Cycle Lemma: In minimal $k$ -vertex-connected graphs, cycles tightly enclose terminal-free components.
Connectivity-Separation Theorem: For triconnected planar graphs, the existence of $k$ vertex/edge-disjoint paths per terminal pair ensures that dynamic programming can localize global connectivity constraints.
Dual Cuts and Laminarity: Planar duality links minimal $k$ -cuts in the primal with cycles in the dual, facilitating the “safe cover” decomposition.
Snug Vertices and Chain Graphs: Snug vertices (degree $k+1$ with unique shores) organize into chain graphs enabling contraction and piecewise DP.

These lemmas guarantee that the decomposition isolates hard global constraints, allowing local subproblems to be solved efficiently.

5. Algorithmic Components and Pseudocode Overviews

Core algorithmic steps for PTAS frameworks include:

Preprocessing: Thinning of candidate links, contraction of chain graphs.
Decomposition: Layering via vertex–face distance, formation of overlapping rings, identification of $k$ -edge-safe covers.
Dynamic Programming: On bounded-treewidth subgraphs, with “connectivity signatures” tracking local path-disjointness.
Merging/Glues: Piecewise-optimal solutions combined using purchased overlap edges and augmentations for boundary cases.
Approximation Guarantee: Costs of overlaps and localized augmentations are bounded by $O(\epsilon \cdot \mathrm{OPT})$ ; remaining local solutions sum to at most OPT.

Illustrative pseudocode for PTAS (Neuwohner et al., 24 Dec 2025):

def PTAS(G, L, ε, k):
    preprocess G to be minimally k-connected
    L_prime = thin_out(L)
    P = compute_snug_paths(G)
    Gp, Lp = contract_paths(G, L_prime, P)
    Up = compute_kplus1_edge_safe_cover(Gp, Lp, ε/6, k)
    Q, boundary_links = select_boundary_paths_and_links(Up, P)
    L_StarStar = boundary_links.union(overlap_links(Up))
    piece_solutions = []
    for i in Up:
        Gi, Li = contract_outside(G, L_StarStar, Up, i)
        OPTi = DP_solve(Gi, Li, k)
        piece_solutions.append(OPTi)
    return L_StarStar.union(*piece_solutions)

Running times are linear in $n$ for fixed $k, \epsilon$ , with singly-exponential (in $k/\epsilon$ ) overheads in dynamic programming (Neuwohner et al., 24 Dec 2025).

6. Trade-Offs: Beyond-Planarity and Higher $k$

When augmentation is permitted in beyond-planar classes (ℓ-planar, ℓ-plane graphs), the strict NP-hardness barrier for planar augmentation at $k\leq 5$ is circumvented. The tight asymptotic tradeoff ℓ = Θ( $k^2$ ) allows every planar input to be augmented to $k$ -connectivity with a local crossing number bounded accordingly (Akitaya et al., 1 Sep 2025). Extension of PTAS principles to these settings utilizes clustering, clique formation in dual subgraphs of size Θ( $k$ ), and matching between clusters.

For strict planar graphs, however, the failure of the brick–boundary cover property and the existence of terminal-free cycles preclude extension of these decompositions to $k > 3$ in certain models (Borradaile et al., 2016). A plausible implication is that approximability results reach a barrier in strictly planar settings for larger $k$ , and PTAS efficacy is restricted by these global topological obstructions.

The presented PTAS/EPTAS results generalize and subsume earlier PTASs for Steiner tree, PTSP, and edge/vertex connectivity in planar graphs. Key methodologies—mortar graph + brick + portal approaches, tree-cycle separator lemmas, and branched decompositions—are central in all these settings (Borradaile et al., 2016). The NP-hardness results preclude the existence of FPTAS for planar $k$ -connectivity augmentation unless P=NP (Neuwohner et al., 24 Dec 2025).

Summary of properties:

Property	PTAS/EPTAS	Planar, $k$ constant	Extension Limit
Running time	$O(n)$ (fixed $k$ )	Yes	Not polynomial for unbounded $k$
Approximation ratio	$(1+\epsilon)$	Yes	FPTAS impossible
PTAS for $k>4$	Only for ℓ-planar	No (strict planar)	NP-hardness barrier
Connection to flips	Triangulation model	Yes	4-Connectivity EPTAS only