operator norm

If a bounded operator stretches vectors, the operator norm records its worst-case stretch — the single largest amplification it ever applies. It answers: across all input directions, what is the most this operator can blow up a length? That worst case is a finite number exactly when the operator is bounded, and it gives the whole space of operators its own geometry.

Precisely: for a bounded operator T from X to Y, the operator norm is ||T|| = sup over x != 0 of ||T x|| / ||x||, equivalently the sup of ||T x|| over all unit vectors ||x|| = 1. It is the smallest C that works in ||T x|| <= C ||x||. Three properties make it a true norm and more: it is a norm, it is submultiplicative (||S T|| <= ||S|| ||T||), and ||I|| = 1.

Why it matters structurally: with this norm, the bounded operators from a normed space into a Banach space form a Banach space themselves. The bounded operators on a single Hilbert space H form a Banach algebra under composition, and with the adjoint operation they become a C*-algebra, satisfying the striking identity ||T^* T|| = ||T||^2. That identity ties the algebra of operators tightly to their norms and underlies the modern spectral theory.

A caveat to keep straight: the operator norm (also called the uniform norm) is one of several useful norms on operators — others include the trace norm and Hilbert-Schmidt norm. Convergence in operator norm is strong; many natural sequences of operators converge in weaker senses (strongly, weakly) but not in operator norm. Always ask which topology on operators is meant.