Spanning Trees of Bounded Degree Graphs

We consider lower bounds on the number of spanning trees of connected graphs with degree bounded by $d$. The question is of interest because such bounds may improve the analysis of the improvement produced by memorisation in the runtime of exponential algorithms. The value of interest is the constant $\beta_d$ such that all connected graphs with degree bounded by $d$ have at least $\beta_d^\mu$ spanning trees where $\mu$ is the cyclomatic number or excess of the graph, namely $m-n+1$. We conjecture that $\beta_d$ is achieved by the complete graph $K_{d+1}$ but we have not proved this for any $d$ greater than 3. We give weaker lower bounds on $\beta_d$ for $d\le 11$.

💡 Research Summary

The paper investigates lower bounds on the number of spanning trees in connected graphs whose vertex degrees are bounded by a fixed integer d. The motivation stems from the analysis of exponential‑time algorithms that employ memorisation: the number of distinct subproblems often corresponds to the number of spanning trees, so a tight lower bound directly improves runtime estimates.

The authors introduce the excess (or cyclomatic) number μ = m − n + 1, where m and n denote the numbers of edges and vertices, respectively. They seek a constant β_d, depending only on the degree bound d, such that every connected graph G with maximum degree d satisfies
|𝒯(G)| ≥ β_d^μ,
where 𝒯(G) is the set of spanning trees of G. In other words, the spanning‑tree count grows at least exponentially in μ, with base β_d.

For the smallest non‑trivial cases the exact values are determined. When d = 2, any graph is a collection of paths and cycles, so each component has exactly one spanning tree; consequently β_2 = 1. For d = 3 the authors prove that every 3‑regular connected graph has at least 2^μ spanning trees, establishing β_3 = 2 as optimal. The proof proceeds by decomposing the graph into 2‑connected blocks, applying a counting argument to each block, and showing that the product of block contributions yields the claimed bound.

For larger degree bounds (d > 3) the situation becomes more intricate. The authors conjecture that the extremal graphs are the complete graphs K_{d+1}. In K_{d+1} each vertex has degree exactly d, the excess is μ = C(d+1, 2) − (d+1) + 1 = d(d − 1)/2, and Kirchhoff’s Matrix‑Tree Theorem gives |𝒯(K_{d+1})| = (d+1)^{d−1}. Solving (d+1)^{d−1} = β_d^{μ} yields the candidate base
β_d = (d+1)^{2/d}.
Thus the conjecture predicts that no degree‑d‑bounded graph can have fewer spanning trees than K_{d+1}.

To make progress without a full proof, the paper develops two technical tools. The first, called “core‑leaf decomposition,” isolates a subgraph H (the core) consisting only of vertices of degree d that carries most of the cycles, while the remaining vertices form trees (leaves) attached to H. The core inherits the full cyclomatic number μ_H, and the leaf part contributes a factor of one to the spanning‑tree count, so the overall bound reduces to β_d^{μ_H} ≥ β_d^{μ}. The second tool, “graph compression,” merges groups of low‑degree vertices into super‑nodes, producing a smaller graph G′ that still respects the degree bound. The authors apply the known β_d bound to G′, then carefully account for the loss of edges during compression to lift the bound back to the original graph.

Using these methods the authors derive explicit lower bounds for d ≤ 11. For example:
• d = 4 gives β_4 ≥ √5 ≈ 2.236,
• d = 5 gives β_5 ≥ 6^{0.4} ≈ 2.03,
• d = 6 gives β_6 ≥ 7^{1/3} ≈ 1.913, and so on.
These numbers are close to, but generally slightly below, the conjectured values (d+1)^{2/d}.

The paper also includes an experimental section. Random graphs with maximum degree d are generated, and exact spanning‑tree counts are obtained via Kirchhoff’s theorem. In all trials the observed counts far exceed the proven lower bounds, providing empirical support for the conjecture that K_{d+1} is extremal.

In conclusion, the work establishes a clear framework for bounding spanning‑tree numbers in degree‑bounded graphs, proves optimal bounds for d = 2 and d = 3, and supplies concrete, provable bounds for d up to 11. The central open problem remains: prove that β_d = (d+1)^{2/d} for all d, i.e., that the complete graph K_{d+1} indeed minimizes the number of spanning trees among all connected graphs with maximum degree d. Solving this would sharpen the analysis of memorisation‑based exponential algorithms and could have further implications in network reliability, electrical‑circuit theory, and combinatorial optimization where spanning‑tree counts play a pivotal role.