The path surgery of the involution argument
DefinitionBunkbedWZSurgeryThe involution used by Gladkov--Pak--Zimin to prove their robustness lemma is built along the path of Eq. [5], which traverses every hyperedge of Hollom's hypergraph exactly once and avoids the transversal vertices. This file records the combinatorial data of that construction.
Along the non-transversal vertices appear in the order
recorded as , and hyperedge is the triple , with the transversal vertex first. So Hollom's hypergraph is a path of six edges, each decorated with one of the three transversal vertices , and each transversal vertex is used exactly twice.
Given a refined configuration and a hyperedge , the involution is
where is the half-reflection and the level swap. Correspondingly is the vertex map that reflects the level of every non-transversal vertex lying strictly beyond along — the predicate — while fixing the transversal vertices and everything at or before .
A subtlety worth recording. Transversal vertices are deliberately not reflected, and cannot be: each serves two hyperedges, one before and one after, so no consistent assignment exists. Consequently is not a graph homomorphism — an edge joining a transversal vertex to a non-transversal one is sent to a non-edge, because only one of its endpoints moves. The construction nonetheless works, because the posts over the transversal set are present in every configuration, so a transversal vertex is always connected to its own counterpart on the other level and the mismatch can be repaired by inserting a post. The correct statement to prove is therefore that carries adjacency to reachability, which suffices to transport connectivity along a walk.
Also provided is , extracting the underlying Wierman--Ziff configuration from a refined one: the first component of a refined state is the lower-level state and the second the upper-level one.
import Definitions.Def_BunkbedWZRefined /-! # The path surgery of the involution argument Section 3.5 of Gladkov-Pak-Zimin, *The bunkbed conjecture is false*, PNAS 122 (2025) e2420725122. The involution used to prove the robustness lemma is built from the path `ρ` of Eq. [5], which traverses every hyperedge of Hollom's hypergraph exactly once and avoids the transversal vertices. Along `ρ` the non-transversal vertices appear in the order `hollomPath`, and hyperedge `i` is the triple `(hollomTr i, hollomPath i, hollomPath (i+1))` -- transversal vertex first. Given a refined configuration `Ψ` and the first hyperedge `e` along `ρ` whose state lies in `Λ²₊`, the involution `phiExt` leaves the hyperedges before `e` alone, applies the half-reflection `eta` at `e`, and applies the level swap `rho` to every hyperedge after `e`. Correspondingly `surgeryMap` reflects the level of every non-transversal vertex lying strictly beyond `e` along `ρ`, fixing the transversal vertices and everything before `e`. Note that transversal vertices are deliberately *not* reflected: each of them serves two hyperedges, one before and one after `e`, so no consistent assignment is possible. The resulting mismatch is harmless because the posts over the transversal set are always present, so a transversal vertex is always connected to its own counterpart on the other level. -/ namespace Bunkbed.Hyper /-- The non-transversal vertices of Hollom's hypergraph, in the order they appear along the path `ρ` of Eq. [5]: `u₁, u₃, u₆, u₅, u₄, u₈, u₁₀`. -/ def hollomPath : Fin 7 → Fin 10 := ![0, 2, 5, 4, 3, 7, 9] /-- The transversal vertex of each hyperedge of Hollom's hypergraph, in the order of `ρ`. -/ def hollomTr : Fin 6 → Fin 10 := ![1, 8, 6, 1, 6, 8] /-- Exchange the two levels of the bunkbed. -/ def lvlSwap : Fin 2 → Fin 2 := fun l => if l = 0 then 1 else 0 /-- The vertices whose level the surgery reflects: the non-transversal vertices lying strictly beyond hyperedge `e` along `ρ`. Transversal vertices are never reflected. -/ def reflected (e : Fin 6) (v : Fin 10) : Bool := decide (∃ j : Fin 7, hollomPath j = v ∧ e.val < j.val) /-- The vertex map of the path surgery. -/ def surgeryMap (e : Fin 6) (x : Fin 10 × Fin 2) : Fin 10 × Fin 2 := if reflected e x.1 then (x.1, lvlSwap x.2) else x /-- The underlying Wierman-Ziff configuration of a refined configuration: the first component of a refined state is the lower-level state, the second the upper-level one. -/ def wzOfExt (Ψ : Fin 6 → WZExt) : Fin 6 × Fin 2 → WZ := fun p => if p.2 = 0 then (WZExt.toPair (Ψ p.1)).1 else (WZExt.toPair (Ψ p.1)).2 /-- The composite involution `φ` of Section 3.5: leave the hyperedges before `e` alone, apply the half-reflection at `e`, and apply the level swap to every hyperedge after `e`. -/ def phiExt (e : Fin 6) (Ψ : Fin 6 → WZExt) : Fin 6 → WZExt := fun h => if h.val < e.val then Ψ h else if h = e then WZExt.eta (Ψ h) else WZExt.rho (Ψ h) end Bunkbed.Hyper