Basis Number of a Graph

• Basis number of a graph is defined as the smallest integer k such that there exists a cycle basis where each edge appears in at most k cycles, reflecting both algebraic and topological characteristics.
• It characterizes graph families with forests having value 0, planar graphs capped at 2, and surfaces or minor-closed classes having bounded or polynomially controlled basis numbers.
• This invariant is essential in linking combinatorial optimization with algebraic graph theory, influencing algorithmic complexity and connections to path algebras.

The basis number of a graph, denoted $\beta(G)$ or $b(G)$, is a combinatorial invariant that quantifies how "evenly" the cycle space of a finite undirected graph can be generated by cycles, with respect to the number of times each edge appears in a basis. Formally, it is the smallest integer $k$ such that there exists a basis of the cycle space of $G$—a minimal generating set, over $\mathbb{F}_2$, of Eulerian subgraphs—in which each edge of $G$ is included in at most $k$ basis elements. This parameter robustly reflects both algebraic and topological properties of the underlying graph and arises naturally in both extremal combinatorics and topological graph theory. The basis number also admits direct generalization and precise connections to invariants of directed graphs and their path algebras, particularly through the theory of Invariant Basis Number for associated algebras.

1. Definition and Core Properties

Let $G=(V,E)$ be a finite simple graph. The cycle space $\mathcal{C}(G)$ is the $\mathbb{F}_2$-vector space of all subsets of $E$ such that, in the subgraph $(V,S)$, every vertex has even degree; equivalently, $\mathcal{C}(G)$ consists of all Eulerian subgraphs of $G$. A cycle basis is any set $\mathcal{B}\subseteq\mathcal{C}(G)$ that forms a basis of this vector space; its cardinality is $\beta_1(G) = |E|-|V|+c(G)$, where $c(G)$ is the number of connected components.

For a given basis $\mathcal{B}$, the ply (or charge) of an edge $e$ is $$\mathrm{ply}(e,\mathcal{B}) = |\{B\in\mathcal{B}: e\in B\}|$$. The basis number is then defined as

$$\beta(G) = \min \{ \max_{e\in E} \mathrm{ply}(e, \mathcal{B}) : \mathcal{B} \text{ is a cycle basis} \}$$

This invariant is robust under common graph operations: subdividing edges does not change the basis number, contractions can only decrease it, and the basis number is subadditive over unions with connected overlap. Key thresholds include: $\beta(G)=0$ if and only if $G$ is a forest; $\beta(G)\le1$ if and only if $G$ is a cactus; $\beta(G)\le2$ characterizes planar graphs (MacLane's theorem) (Bazargani et al., 2024, Lehner et al., 2024, Geniet et al., 8 Jan 2026).

2. Extremal Cases and Planarity Criteria

Mac Lane's planarity criterion asserts that $G$ is planar if and only if $\beta(G)\le2$, i.e., there exists a cycle basis where each edge lies in at most two cycles. This connects the basis number to classical notions of surface embeddings and planar duality.

Beyond planarity, the basis number is tightly controlled for families of graphs embeddable on fixed surfaces. For example, every non-planar graph embeddable on a surface of Euler characteristic $0$ (the torus or Klein bottle) satisfies $\beta(G)=3$ (Lehner et al., 2024). More generally, for graphs embedded on a surface of genus $g$ (orientable or non-orientable), $\beta(G) = O(\log^2 g)$ (Lehner et al., 2024). In minor-closed classes, the basis number is bounded as a function of the excluded minor: every $H$-minor-free graph satisfies $\beta(G)=O(|H|^c)$ for a universal constant $c$ (Geniet et al., 8 Jan 2026).

For families defined by structural constraints, such as pathwidth or treewidth, sharp polynomial upper bounds are available. Any graph $G$ with pathwidth $k$ has $\beta(G)\le4k$ (Miraftab et al., 20 Jan 2026). More generally, if $G$ has treewidth $k$, then $\beta(G)=O(k^5)$ (Miraftab et al., 20 Jan 2026, Geniet et al., 8 Jan 2026), and for arbitrary $K_t$-minor-free graphs, the bound is polynomial in $t$ (Miraftab et al., 20 Jan 2026, Geniet et al., 8 Jan 2026).

3. Structural Theorems and Graph Classes

The basis number exhibits unbounded behavior in certain topological classes: for instance, the family of 1-planar graphs (those drawable in the plane with at most one crossing per edge) has unbounded basis number (Bazargani et al., 2024). Nevertheless, several important subclasses admit uniform bounds. Every 1-planar graph whose "skeleton" (the subgraph of uncrossed edges in a drawing) is connected obeys $\beta(G)\le4$, and the same is true for locally maximal and "poppy" 1-planar graphs. For 2-connected "full-crossing" or optimal 1-planar graphs, $\beta(G)\le3$ (Bazargani et al., 2024).

For minor-closed families, the basis number is governed by the Graph Minor Structure Theorem. In particular, almost-embeddable pieces—graphs decomposable into a bounded genus-embedded core plus bounded pathwidth vortices—have bounded basis number, and clique-sums across small adhesions do not inflate $\beta(G)$ beyond polynomial bounds in the relevant parameters (Geniet et al., 8 Jan 2026).

A summary table of key structural bounds is as follows:

Graph Class	Basis Number Bound	Reference
Forests	$\beta(G)=0$	(Miraftab et al., 20 Jan 2026)
Planar	$\beta(G)\le2$	(Bazargani et al., 2024)
Genus-$g$ surface	$\beta(G)=O(\log^2g)$	(Lehner et al., 2024)
Torus/Klein bottle	$\beta(G)=3$	(Lehner et al., 2024)
Pathwidth $k$	$\beta(G)\le4k$	(Miraftab et al., 20 Jan 2026)
Treewidth $k$	$\beta(G)=O(k^5)$	(Miraftab et al., 20 Jan 2026)
$K_t$-minor-free	$\beta(G)=O(t^c)$	(Geniet et al., 8 Jan 2026)
1-planar (general)	unbounded	(Bazargani et al., 2024)
1-planar (connected skeleton)	$\beta(G)\le4$	(Bazargani et al., 2024)
Full-crossing 1-planar	$\beta(G)\le3$	(Bazargani et al., 2024)

4. Methodologies and Proof Strategies

The computation and bounding of $\beta(G)$ exploit a blend of algebraic, topological, and combinatorial techniques. In the planar and surface-embedded settings, the approach hinges on face-cycle bases and careful manipulation of fundamental cycles relative to an appropriate spanning tree. The proof that $\beta(G)=3$ for toroidal and Klein bottle embeddings utilizes a replacement lemma which reduces edge coverage via structured combinations of face cycles and fundamental cycles (Lehner et al., 2024).

For bounded-width decompositions, inductive constructions on path or tree decompositions allow for local control of basis ply and global bounds via covering and merging lemmas. The analysis of the ply increase under successive augmentations is crucial (Miraftab et al., 20 Jan 2026). In minor-closed classes, the gluing of pieces via bounded-adhesion tree decompositions and the Simon Factorization Forest Theorem enables the preservation of basis number bounds across composition operations (Geniet et al., 8 Jan 2026).

Algebraic counterparts, notably in Leavitt and Cohn path algebras, interpret the basis number as the well-defined rank of free modules—provided the algebra satisfies the Invariant Basis Number (IBN) property, which is characterized by explicit linear algebraic criteria involving the graph's incidence matrix (Nam et al., 2016, Abrams et al., 2013). For instance, $L_K(E)$ has IBN if and only if the rank jump condition $$\mathrm{rank}_\mathbb{Q}(M) < \mathrm{rank}_\mathbb{Q}(M')$$ holds, where $M = A_E^T - J_E$ is derived from the graph's incidence matrix.

5. Illustrative Examples and Special Cases

• Forests: $\beta(G)=0$, as there are no cycles.
• Cacti: $\beta(G)\le1$, as each edge belongs to at most one independent cycle.
• Planar graphs: Every planar $G$ has a $2$-basis; e.g., the facial cycles of a plane embedding suffice.
• Klein bottle and torus: For any non-planar $G$ embedded on these surfaces, $\beta(G)=3$ is achievable via explicit face and fundamental cycles (Lehner et al., 2024).
• 1-planar, full-crossing: All optimal 1-planar graphs (achieving $4n-8$ edges) admit a $3$-basis by merging local $K_4$ "poppy" substructures with global face cycles (Bazargani et al., 2024).

Combinatorial operations such as subdivision and contraction preserve or reduce the basis number, supporting robust inductive and constructive approaches for bounding $\beta(G)$ across families.

6. Connections to Algebraic Graph Invariants

In the context of Leavitt and Cohn path algebras associated to finite (directed) graphs, the Invariant Basis Number relates directly to a "basis number" for the module category. For $L_K(E)$ or $C_K(E)$, IBN assures that the rank of any free module is uniquely determined—mirroring the uniqueness of basis cardinalities in the combinatorial setting. The IBN property can be algorithmically detected via a matrix rank condition on graph incidence data: $L_K(E)$ has IBN if and only if the extended matrix $[A_E^T - J_E | b]$ has strictly higher rank than $A_E^T - J_E$ over $\mathbb{Q}$ (Nam et al., 2016).

This algebraic perspective underscores the interplay between the combinatorial basis number and module-theoretic invariants and motivates further connections between graph structures and algebraic properties of associated path algebras (Abrams et al., 2013).

7. Open Problems and Computational Complexity

Several fundamental problems remain open:

• The algorithmic complexity of computing $\beta(G)$, even for 1-planar graphs, is unknown; the problem is conjectured to be NP-hard.
• The extremal growth of the basis number (e.g., whether $b(G)=O(\log n)$ for 1-planar graphs with $n$ vertices) is not settled.
• Characterizing minimal (in terms of vertex count) 1-planar graphs with $\beta(G)\ge4$ remains open.
• The tightness of polynomial bounds in minor-closed families and the explicit dependence of the exponent $c$ in $\beta(G)=O(|H|^c)$ are areas of ongoing analysis (Bazargani et al., 2024, Lehner et al., 2024, Geniet et al., 8 Jan 2026).

These questions interface directly with current research in structural graph theory, algebraic graph invariants, and computational graph algorithms.