← All papers
First page of A Counterexample to Wormald's Conjecture

A Counterexample to Wormald's Conjecture

James Alexander Schreib

math.CO Sep 7, 2026 · v1
The 16-vertex counterexample and the infinite family of counterexamples are formalized in Lean 4 with Mathlib, using kernel-checked certificates and no custom axioms.
We give a counterexample to Wormald's conjecture. A 16-vertex cubic graph, the disjoint union of $K_{3,3}$ with a 10-vertex bridged graph built from two subdivided copies of $K_4$, has no partition of its edges into two isomorphic spanning linear forests. The obstruction is parity: any such partition would force every monochromatic component to have even order while requiring a five-vertex side of a bridge to be partitioned into such components. Adjoining copies of $K_4$ gives counterexamples in every order $16+4t$. The counterexample is disconnected and leaves the connected case unresolved. The full family is verified in Lean without custom axioms.

Wormald conjectured that every cubic graph of order divisible by four has an edge partition into two isomorphic spanning linear forests. The question is whether this holds for all such cubic graphs, including disconnected ones.

The witness graph is the disjoint union of K_{3,3} with a 10-vertex bridged cubic graph built from two subdivided copies of K_4. Degree counting and the complement structure on K_{3,3} force every monochromatic component to have even order. A parity argument on a five-vertex side of the bridge then gives a contradiction. In Lean, the component inventories of all 512 edge subsets of K_{3,3} are checked with a kernel-checked certificate checker, and the inventory and parity lemmas are proved formally.

The 16-vertex graph is a counterexample, and adjoining copies of K_4 gives counterexamples of every order 16+4t. The connected case remains open. Both the concrete theorem and the family theorem are machine-checked in Lean and depend only on the standard axioms (propext, Classical.choice, Quot.sound).