Local Hurwitz cancellation in a free group
ProvedBraidsLinksMCG.hurwitz_local_cancellationA purely free-group cancellation statement; no braid group appears.
Let be elements of the free group , each a conjugate of a generator,
whose product in order is the whole word:
Assume the tuple is not already the trivial one up to relabelling, i.e. it is not the case that every is itself a generator. Then some adjacent pair admits a length-decreasing conjugation: either
the norm being the length of the reduced word.
What this is. These are exactly the two Hurwitz moves at position ,
each of which preserves both the ordered product and the property of being a tuple of conjugates of generators. The claim is that as long as the tuple is not the base tuple , one of the moves strictly decreases . Since one move changes only two entries, and the entry merely shifts position in the first case and in the second, the change in the total is exactly the difference displayed above.
Why it is plausible and where the content lies. Normalise each so that is reduced as written; then and the total is , while the product has length exactly . So the amount of cancellation occurring in is forced to be as large as it can be, and the whole difficulty is to convert that global cancellation into a statement about one adjacent pair. The hypothesis that not every is a generator says precisely that some is non-trivial, i.e. that the total exceeds its minimum value .
The statement is the combinatorial heart of Artin's characterization of the braid automorphisms of a free group, isolated from the braid group: it is what makes the descent in BraidsLinksMCG.artin_hurwitz_norm_reduction terminate, and through that it is the last non-topological obstruction in this mission. It involves no presented group, no group action, and no topology --- only reduced words in a free group.
Note that the conclusion is genuinely a disjunction over both directions of the move. The first alternative alone is false: a tuple can be at a local minimum for all positive moves while a negative move still shortens it.
import Mathlib import Definitions.Def_BraidsLinksMCG_ArtinBraidGroup import Definitions.Def_BraidsLinksMCG_ArtinEndo
namespace BraidsLinksMCG
theorem hurwitz_local_cancellation (n : ℕ)
(w : Fin n → FreeGroup (Fin n))
(mu : Equiv.Perm (Fin n)) (A : Fin n → FreeGroup (Fin n))
(hw : ∀ i : Fin n, w i = A i * FreeGroup.of (mu i) * (A i)⁻¹)
(hprod : (List.ofFn w).prod = freeWordProd n)
(hnotperm : ¬ ∃ nu : Equiv.Perm (Fin n), ∀ i : Fin n, w i = FreeGroup.of (nu i)) :
∃ a b : Fin n, (b : ℕ) = (a : ℕ) + 1 ∧
(FreeGroup.norm (w a * w b * (w a)⁻¹) < FreeGroup.norm (w b) ∨
FreeGroup.norm ((w b)⁻¹ * w a * w b) < FreeGroup.norm (w a)) := by sorry
end BraidsLinksMCG