Hamilton Decompositions of the Directed 3-Torus: A Return-Map and Odometer View
This lightning talk explores a breakthrough in graph theory: proving that every directed three-dimensional torus can be partitioned into exactly three arc-disjoint Hamiltonian cycles. The presentation reveals how the authors reduced this complex problem from a million-vertex question to a thousand-vertex one using return maps and odometer dynamics, exposed a fundamental parity obstruction that splits odd and even cases into entirely different worlds, and formalized the entire proof in Lean 4 for computational verification.Script
Take a graph with a million vertices and a million directed edges. Now prove that you can trace exactly three paths through it, each visiting every vertex once, never overlapping. The authors just solved this for every directed three-dimensional torus, and their method is a masterclass in dimensional reduction.
The directed 3-torus is built by stacking directed cycles in three dimensions. Every vertex has exactly three outgoing arcs, one in each generator direction. The question: can you always partition all arcs into exactly three Hamiltonian cycles that never share an edge?
The key insight transforms an intractable global problem into a local one.
Every arc in the torus increases a sum coordinate by one, so after exactly m steps, you're guaranteed to return to the starting layer. This creates a Poincaré-like section: instead of tracking paths through a million vertices, you only need to understand how m-step returns behave on a thousand-vertex plane.
Here's where it gets surprising. For odd m, you can reach a Hamilton decomposition through clever recoloring moves called Kempe swaps, and the dynamics resemble a simple two-dimensional clock with carry. But for even m, a sign-product invariant proves that Kempe swaps can never work. The authors had to build an entirely different construction from scratch.
The proof isn't just rigorous on paper. The entire argument is formalized in Lean 4, with computational verification for finite cases. The return-map technique promises to unlock higher-dimensional cases, and the parity divide hints at fundamental limits in how graphs can be recolored and decomposed.
When a million-vertex problem reduces to a thousand-vertex one, and a parity invariant splits reality into two incompatible worlds, you've found something structural. Visit EmergentMind.com to explore more research and create your own videos.
