Newton's method over an adically complete ring #
Let R be a commutative ring and π ∈ R an element for which R is π-adically complete
(IsAdicComplete (Ideal.span {π}) R). Let A : (ι → R) → (ι → R) be a map on a finite free
module, M a square matrix over R whose determinant is a unit, and u₀ a point with
A u₀ ≡ 0 mod π. Suppose M is a uniform linearisation of A on the residue class of u₀:
for every k ≥ 1 and all u, u' in that class with u' ≡ u mod π^k,
A u' ≡ A u + M (u' - u) mod π^(k+1).
Then A has exactly one zero in the residue class of u₀
(TauCeti.IsAdicComplete.existsUnique_eq_zero_of_isUnit_det).
This is a multivariable form of Hensel's lemma with an integral linearisation. No polynomiality of
A is assumed: the linearisation hypothesis plays the role of the derivative, and the unit
determinant of M that of the nonvanishing of the Jacobian modulo π. Mathlib's
Mathlib.NumberTheory.Padics.Hensel is the one-variable polynomial case over ℤ_p. The theorem
supplies the canonical character of a Demushkin group: its values on the generators are the zero of
the map recording the values of the crossed homomorphisms on the relator.
Main results #
TauCeti.IsAdicComplete.existsUnique_eq_zero_of_isUnit_det: Newton's method, the existence and uniqueness of the zero.
Newton's method over a π-adically complete ring. Let A : (ι → R) → (ι → R), let M be
a matrix with unit determinant, and let u₀ satisfy A u₀ ≡ 0 mod π. If M linearises A
uniformly on the residue class of u₀, in the sense that A u' ≡ A u + M (u' - u) mod π^(k+1)
whenever u, u' lie in that class and u' ≡ u mod π^k with k ≥ 1, then A has exactly one zero
in the residue class of u₀.