Units and the Jacobson radical of normed rings #
In a normed ring with summable geometric series, Ring.inverse is continuous at every unit
(NormedRing.inverse_continuousAt), but it jumps to 0 at the boundary of the open set of units.
When the nonunits also form an open set, as in a complete discretely valued ring such as ℤ_p,
there is no such boundary and Ring.inverse is continuous everywhere.
The Neumann series also gives explicit bounds: ‖(1 - x)⁻¹‖ ≤ ‖1‖ - 1 + (1 - ‖x‖)⁻¹ when
‖x‖ < 1, and, when ‖1‖ = 1, ‖(u + t)⁻¹‖ ≤ ‖u⁻¹‖ / (1 - ‖u⁻¹‖ * ‖t‖) for a unit u with
‖u⁻¹‖ * ‖t‖ < 1; in particular (1 - w) * u + d is a unit with
‖((1 - w) * u + d)⁻¹‖ ≤ M / (1 - ‖w‖ - M * ‖d‖) when ‖u⁻¹‖ ≤ M, ‖w‖ < 1 and
M * ‖d‖ < 1 - ‖w‖.
Mathlib records these inverses only through asymptotic O-bounds such as
NormedRing.inverse_add_norm; the explicit constants are needed for inductive constructions
that perturb a unit infinitely often, such as Cohen–Hewitt factorization.
The Jacobson radical is closed, as an intersection of maximal left ideals, using
Mathlib's Ring.jacobson_eq_sInf_isMaximal and Ideal.IsMaximal.isClosed.
See Dales, Banach Algebras and Automatic Continuity, §§1.3–1.5.
If the nonunits of a normed ring with summable geometric series form an open set, then
Ring.inverse is continuous: it is continuous at every unit, and locally zero at every nonunit.
The Neumann-series bound for the inverse of 1 - x when ‖x‖ < 1. The identity need not have
norm one; when it does, the bound reads ‖(1 - x)⁻¹‖ ≤ (1 - ‖x‖)⁻¹.
In a normed ring with ‖1‖ = 1 and summable geometric series, perturbing a unit u by t
with ‖u⁻¹‖ * ‖t‖ < 1 gives an element whose inverse has norm at most
‖u⁻¹‖ / (1 - ‖u⁻¹‖ * ‖t‖). The perturbed element is a unit by Units.add.
Perturbing a unit a with ‖a⁻¹‖ ≤ M to (1 - w) * a + d, where ‖w‖ < 1 and
M * ‖d‖ < 1 - ‖w‖, gives a unit whose inverse has norm at most M / (1 - ‖w‖ - M * ‖d‖).
The Jacobson radical of a normed ring with summable geometric series is closed.