the Hille-Yosida theorem
/ HIL-uh yo-SHEE-dah /
The Hille-Yosida theorem is the foundational existence-and-characterization result of semigroup theory: it answers exactly which (possibly unbounded) operators A are the generator of a strongly continuous contraction semigroup. This is the converse direction to differentiating P_t — given a candidate local rule A, can we integrate it into a global flow P_t = e^(tA)? — and it is what licenses solving evolution equations and constructing Markov processes from their generators.
Statement (contraction case): a linear operator A on a Banach space X generates a C_0 contraction semigroup if and only if (i) A is densely defined and closed; (ii) every lambda > 0 belongs to the resolvent set, i.e. (lambda I - A) is invertible with bounded inverse R(lambda) = (lambda I - A)^(-1); and (iii) the resolvent satisfies the bound ||R(lambda)|| <= 1/lambda for all lambda > 0. The deep content is the construction of the semigroup from the resolvent. Yosida's idea is the Yosida approximation A_lambda = lambda A R(lambda) = lambda^2 R(lambda) - lambda I, which is a BOUNDED operator approximating A; for bounded operators e^(t A_lambda) is defined by the ordinary power series, and one shows e^(t A_lambda) converges strongly as lambda -> infinity to a limit semigroup whose generator is A. The resolvent itself is the Laplace transform of the semigroup, R(lambda) f = integral_0^infinity e^(-lambda t) P_t f dt, which is the formula tying the two objects together.
Why it matters: Hille-Yosida turns 'I have a generator' into 'I have a process/flow', so it underlies the construction of Feller processes, diffusions and solutions of the Kolmogorov equations. For Markov semigroups one adds a positivity condition (the resolvent maps non-negative functions to non-negative functions, plus P_t 1 = 1) to get the Lumer-Phillips / positive-maximum-principle version. Crucial honesty: the conditions are NOT cosmetic. Verifying the range condition that (lambda I - A) is onto is usually the hard part and is exactly where the existence of the process can fail; the bound ||R(lambda)|| <= 1/lambda is the contraction (sub-Markov) requirement, and weakening it to ||R(lambda)|| <= M/(lambda - omega) gives the general (non-contraction) Hille-Yosida-Phillips theorem for semigroups with growth ||P_t|| <= M e^(omega t).
Take A = (1/2) d^2/dx^2 on C_0(R) with domain the C^2 functions vanishing at infinity with vanishing second derivative. One checks (lambda I - A) f = g is solvable with the resolvent R(lambda) being convolution against (1/sqrt(2 lambda)) e^(-sqrt(2 lambda) |x|), and ||R(lambda)|| <= 1/lambda. Hille-Yosida then constructs the heat semigroup — i.e. Brownian motion.
From the resolvent bound on (lambda I - A), Hille-Yosida manufactures the whole semigroup P_t = e^(tA).
The hard, often-failing hypothesis is the range condition (lambda I - A onto) — exactly where existence/uniqueness of the process can break down; the norm bound is the contraction (sub-Markov) requirement.