Barnette's Conjecture and Hamiltonian Graphs

• Barnette’s Conjecture is the assertion that every cubic, 3-connected, planar, bipartite graph is Hamiltonian, equivalently stating that every simple even plane triangulation can be partitioned into two induced trees.
• The conjecture serves as a pivotal connection among Hamiltonian cycles, planar duality, and matching theory, and has driven research in computational verification and complexity analysis.
• Its extensive literature features equivalent formulations, structural reductions, and algorithmic strategies, with computer-assisted verifications confirming properties for graphs up to 90 vertices.

Barnette’s Conjecture is the assertion that every cubic, $3$-connected, planar, bipartite graph is Hamiltonian (Arts et al., 2013). In dual form, it is equivalent to the statement that every simple even plane triangulation admits a partition of its vertex set into two induced trees (Florek, 2012). The conjecture lies at the intersection of Hamiltonian graph theory, planar duality, matching theory, and algorithmic complexity, and it has generated a substantial literature of equivalent formulations, structural reductions, infinite Hamiltonian subclasses, and large-scale computer verification (Alt et al., 2013, Gorsky et al., 2022, Brinkmann et al., 2021).

1. Statement, terminology, and duality

Barnette’s Conjecture is usually stated as follows: every cubic, $3$-connected, planar, bipartite graph contains a Hamiltonian cycle (Arts et al., 2013). In the supplied literature, such graphs are also called Barnette graphs (Gh. et al., 2022). Here, “cubic” means every vertex has degree three; “planar” means the graph admits a crossing-free embedding in the plane; “bipartite” means the vertex set can be partitioned into two classes so that every edge joins opposite classes; and “$3$-connected” means deleting any two vertices leaves the graph connected (Brinkmann et al., 2021, Arts et al., 2013).

A standard dual formulation replaces the primal cubic bipartite plane graph by an even plane triangulation. If $G$ is a planar triangulation all of whose vertices have even degree, then its dual $G^*$ is cubic, planar, and bipartite; conversely, the dual of a cubic planar bipartite graph is an Eulerian planar triangulation (Alt et al., 2013). In this language, Barnette’s Conjecture becomes: if $G$ is an Eulerian planar triangulation, then $G^*$ is Hamiltonian (Alt et al., 2013). Equivalently, every simple even plane triangulation admits a partition of its vertex set into two subsets so that each induces a tree (Florek, 2012).

This duality is not merely formal. Several later approaches proceed entirely on the triangulation side, where $3$-colourings, induced forests, and face-structure become the central objects (Florek, 2012, Alt et al., 2013, Florek, 2020). A recurrent theme is that Hamiltonicity in the cubic graph is equivalent to an acyclic covering structure in the dual triangulation.

2. Equivalent formulations and structural reductions

A basic equivalence used repeatedly in the literature is the permeating-subtree criterion. For a planar triangulation $G$, the following are equivalent: $G^*$ has a Hamiltonian cycle; $3$0 contains an induced subtree that meets every face; and there exist two disjoint induced subtrees of $3$1 whose union is $3$2 and each of which meets every face (Alt et al., 2013). This criterion underlies colouring-based sufficient conditions and several dual proofs.

Kelmans’ equivalences identify Barnette’s Conjecture with stronger-looking edge-forcing statements. One formulation states that the following are equivalent: every cubic, $3$3-connected, planar, bipartite graph is Hamiltonian; for every such graph $3$4 and every pair of distinct edges $3$5 on the same face, there is a Hamiltonian cycle containing $3$6 but avoiding $3$7; every cyclically-$3$8-edge-connected cubic, $3$9-connected, planar, bipartite graph is Hamiltonian; and, in that cyclically-$3$0-edge-connected class, every two edges $3$1 on a common face lie on a common Hamiltonian cycle (Alt et al., 2013). These equivalences justify restricting attention to cyclically-$3$2-edge-connected instances without loss of generality (Alt et al., 2013).

A matching-theoretic reformulation pushes the conjecture beyond planarity. Using Pfaffianity and tight-cut decomposition, one paper states that Barnette’s Conjecture holds if and only if all cubic, $3$3-connected, Pfaffian, bipartite graphs are Hamiltonian (Gorsky et al., 2022). The same work shows that Barnette’s Conjecture holds if and only if every cubic, planar brace is Hamiltonian (Gorsky et al., 2022). Here a brace is a bipartite matching covered graph that is either isomorphic to $3$4 or is $3$5-extendable (Gorsky et al., 2022). This reduction isolates braces as the indecomposable matching-theoretic building blocks relevant to Barnette’s problem.

Further strengthened Hamiltonicity properties also admit brace reductions. For cubic, $3$6-connected, bipartite graphs, the paper (Gorsky et al., 2022) studies $3$7-Hamiltonicity, the $3$8-property, $3$9-Hamiltonicity, and the $G$0-property, and states that for each of these properties a graph has the property if and only if all its braces have the property (Gorsky et al., 2022). This places Barnette-type Hamiltonicity within a broader family of edge- and path-forcing conditions.

3. Sufficient conditions and infinite Hamiltonian subclasses

A substantial part of the literature establishes Hamiltonicity under additional structural hypotheses. One dual sufficient condition begins with a planar triangulation whose vertices are coloured blue and red so that every face is incident with at least one blue vertex, there is no cycle entirely in blue vertices, and every cycle in the red-induced subgraph contains a vertex of degree at most $G$1 in the triangulation. Under these assumptions, the dual graph is Hamiltonian (Alt et al., 2013). A corollary applies this to Eulerian planar triangulations with a proper blue-red-green colouring: if every red-green cycle contains a vertex of degree $G$2, then the dual is Hamiltonian (Alt et al., 2013).

Florek’s work gives several dual sufficient conditions in terms of the degree-$G$3 “big” vertices of a simple even plane triangulation. If $G$4 and $G$5 are acyclic, then for every path $G$6 there is a partition of the vertex set into two induced trees so that one contains the edge $G$7 and avoids $G$8, and for every path $G$9 with $G^*$0 and $G^*$1 of the same colour there is such a partition with one tree containing the whole path $G^*$2 (Florek, 2012). By duality, this yields Hamiltonicity in the corresponding primal Barnette graphs (Florek, 2012).

A later sufficient condition is phrased directly in terms of faces of the cubic plane graph. Let $G^*$3 be a cubic, $G^*$4-connected, plane, bipartite graph, and call a face small if it has exactly four edges and big if it has at least six. If no face has more than four big neighbours, then $G^*$5 is Hamiltonian (Florek, 2023). Under the additional assumption that every vertex is incident with both a small and a big face, the same paper states that $G^*$6 has at least

$G^*$7

where $G^*$8 is the number of big faces and $G^*$9 is the maximum face-size among the big faces (Florek, 2023).

Other infinite subclasses arise from face-size restrictions. Goodey’s classical result states that if every big face has size exactly $G$0, then the graph is Hamiltonian (Florek, 2023). A much stronger recent theorem states that every finite, simple, cubic, bipartite, planar, connected graph with all faces of size at most $G$1 is Hamiltonian; as an immediate corollary, every Barnette graph with faces of size at most $G$2 is Hamiltonian (Schnieders, 5 Aug 2025). The proof uses a minimal-counterexample strategy, local graph substitutions, the identity

$G$3

and a computational verification of finitely many substitution cases (Schnieders, 5 Aug 2025).

Further infinite subclasses are defined by additional decomposition structure. Annular decomposable Barnette graphs with non-singular sequences of ring annuli are Hamiltonian (Bej, 2020). Likewise, the leapfrog extension of a cyclically $G$4-edge-connected bipartite cubic planar graph is Hamiltonian (Gh. et al., 2018). Another quantitative result concerns the subclass $G$5 of cubic, $G$6-connected, bipartite plane graphs admitting a $G$7-factor consisting only of facial $G$8-cycles: every such graph has at least

$G$9

different Hamilton cycles (Florek, 2018).

4. Computational verification and finite-range results

Computer-assisted verification has produced the strongest finite-range evidence recorded in the supplied literature. A 2021 paper reports a complete computer-assisted verification of Barnette’s Conjecture for all $G^*$0-connected planar bipartite cubic graphs up to $G^*$1 vertices (Brinkmann et al., 2021). The central theorem there states:

• if $G^*$2, then every such graph is Hamiltonian;
• if $G^*$3, then every edge lies on at least one Hamiltonian cycle;
• if $G^*$4, then for any two distinct edges $G^*$5, there exists a Hamiltonian cycle containing $G^*$6 and avoiding $G^*$7 (Brinkmann et al., 2021).

The computation exhaustively enumerated all $G^*$8-connected planar cubic graphs of order $G^*$9 with plantri, then filtered the bipartite instances (Brinkmann et al., 2021). Hamiltonicity testing was carried out by cubhamg, a complete backtracking algorithm optimized for cubic graphs (Brinkmann et al., 2021). Each edge is labelled YES, NO, or UNDECIDED, and a propagation routine applies local degree constraints; after propagation stalls, the algorithm branches on an undecided edge (Brinkmann et al., 2021).

The computational scale is explicitly documented. The enumeration of all $3$0-connected planar cubic graphs up to $3$1 took approximately $3$2 CPU·years, while the two-edge property up to $3$3 required about $3$4 CPU·years, plus a further $3$5 CPU·years for exceptional cases of order $3$6 and $3$7; overall, the paper reports about $3$8 CPU·years for the strongest part and about $3$9 CPU·years for the remaining parts (Brinkmann et al., 2021). The same paper states that these bounds improve the previous record of $G$0 (Brinkmann et al., 2021).

Earlier computational work had established Hamiltonicity up to $G$1 vertices using the two reduction operations $G$2 and $G$3 to generate all cubic, $G$4-connected, bipartite, planar graphs from the unique $G$5-vertex base polyhedron $G$6 (Arts et al., 2013). The later matching-theoretic work revisits this generation framework and augments it with dynamic tracking of non-trivial tight cuts, so that brace and non-brace cases can be distinguished during generation with $G$7 space overhead and $G$8 update time per expansion (Gorsky et al., 2022).

The most recent face-size-bounded proof also has a large computational component. The theorem for cubic, bipartite, planar, connected graphs with faces of size at most $G$9 requires verification of $G^*$0 substitution cases in total, implemented in SageMath (Schnieders, 5 Aug 2025). This is presented as a finite reducibility verification analogous in spirit to local-configuration methods (Schnieders, 5 Aug 2025).

5. Algorithmic formulations and complexity

Barnette’s Conjecture has a pronounced algorithmic aspect because it lies near a complexity boundary. One survey states that deciding whether a given cubic, $G^*$1-connected, planar graph has a Hamiltonian cycle is NP-complete, and likewise for variants where one omits either bipartiteness or planarity (Arts et al., 2013). Another line of work studies Hamiltonicity in planar cubic graphs possessing a facial $G^*$2-factor via quotient graphs and spanning trees of faces (Gh. et al., 2022, Gh. et al., 2018).

If $G^*$3 is a planar cubic graph with a facial $G^*$4-factor $G^*$5, then contracting each cycle of $G^*$6 yields a quotient graph $G^*$7 (Gh. et al., 2022). In this setting, Hamiltonian cycles in $G^*$8 correspond to quasi spanning trees of faces in $G^*$9 of a prescribed inside-outside type (Gh. et al., 2022, Gh. et al., 2018). This leads to polynomial-time solvable subcases. When the relevant family of faces in $3$00 consists only of digons and triangles, deciding whether $3$01 admits a spanning tree of faces reduces in polynomial time to the Spanning-Tree Parity Problem, which is solvable in $3$02 time (Gh. et al., 2022).

The negative side is equally prominent. Even under strong restrictions on the quotient graph, deciding whether a spanning tree of faces exists can be NP-complete (Gh. et al., 2022). Most strikingly, one paper proves the conditional statement that if Barnette’s Conjecture is false, then Hamiltonicity in $3$03-connected planar cubic bipartite graphs is NP-complete (Gh. et al., 2022). A related earlier paper proves the same conditional NP-completeness statement using forced-edge gadgets derived from a minimal non-Hamiltonian Barnette graph (Gh. et al., 2018).

This conditional dichotomy makes the conjecture unusual. If true, Hamiltonicity on Barnette graphs is trivial as a decision problem; if false, the cited work indicates NP-completeness (Gh. et al., 2022, Gh. et al., 2018). Several authors therefore treat Barnette’s Conjecture as identifying a particularly tight frontier between tractable and NP-hard Hamiltonicity regimes (Arts et al., 2013, Gh. et al., 2022).

6. Status, scope, and disputed claims in the literature

Most of the supplied literature treats Barnette’s Conjecture as an open problem. The 2021 computational verification describes it as “one of the most celebrated open problems in the theory of Hamiltonian cycles” and proves only that no counterexample exists up to $3$04 vertices (Brinkmann et al., 2021). The 2022 matching-theoretic paper again presents it as a conjecture and develops equivalent formulations rather than a resolution (Gorsky et al., 2022). The 2025 face-size-$3$05 result is explicitly framed as a substantial strengthening of earlier partial results rather than a complete solution (Schnieders, 5 Aug 2025).

At the same time, the supplied record contains two preprints making stronger claims. The abstract of “Cyclic Subsets and Barnette’s Conjecture” states that cyclic subsets are used to construct an inductive proof of Barnette’s long-standing conjecture (Clarke, 2013). Separately, “A computer-assisted proof of Barnette-Goodey conjecture: Not only fullerene graphs are Hamiltonian” claims a proof that every $3$06-connected planar cubic graph with face-sizes at most $3$07 is Hamiltonian, and notes that Barnette’s original conjecture is the special bipartite case in which all faces have size exactly $3$08 or $3$09 (Kardoš, 2014). Since later papers in the supplied corpus continue to regard Barnette’s Conjecture as unresolved, these claims coexist with a continuing open-problem literature rather than with a settled consensus.

Within that ongoing literature, the current shape of the subject is clear. The conjecture has many equivalent formulations: induced-tree partitions in even plane triangulations (Florek, 2012), permeating subtrees (Alt et al., 2013), edge-forcing variants due to Kelmans (Alt et al., 2013), quasi spanning trees of faces (Gh. et al., 2022), and Hamiltonicity of cubic Pfaffian bipartite graphs or cubic planar braces (Gorsky et al., 2022). It also has a large catalogue of positive cases: bounded order up to $3$10 vertices (Brinkmann et al., 2021), bounded face-size up to $3$11 (Schnieders, 5 Aug 2025), the four-big-neighbours condition (Florek, 2023), multi-$3$12-cycle dual conditions (Florek, 2020), annular decomposable subclasses (Bej, 2020), leapfrog extensions (Gh. et al., 2018), and the facial-$3$13-cycle subclass $3$14 with exponentially many Hamiltonian cycles (Florek, 2018).

A plausible implication is that the conjecture has resisted resolution not because of a lack of reformulations, but because each reformulation isolates a different obstruction class without eliminating all of them simultaneously. The literature supplied here therefore presents Barnette’s Conjecture less as a single isolated statement than as a nexus connecting planar duality, cyclic edge-connectivity, matching theory, face-structure, and computational reducibility.