the Hille-Yosida theorem
/ HILL-uh yoh-SHEE-dah /
Here is the question the theorem settles. You have an abstract Cauchy problem du/dt = A u, u(0) = u_0, and you want to know: does it have a unique, well-behaved solution for every initial state — equivalently, does the operator A generate a C_0 semigroup e^{tA}? You would love a checklist you can verify directly on A, without ever constructing the semigroup. The Hille-Yosida theorem is exactly that checklist: a clean necessary-and-sufficient condition, stated entirely in terms of the resolvent of A.
The cleanest version (for contraction semigroups, those with norm of T(t) <= 1) reads: a densely defined closed operator A generates a C_0 semigroup of contractions if and only if every real lambda > 0 lies in the resolvent set of A and the resolvent satisfies norm of (lambda I - A)^{-1} <= 1/lambda. The general statement allows growth norm of T(t) <= M e^{omega t}, and then requires the powers of the resolvent to be bounded, norm of (lambda I - A)^{-n} <= M/(lambda - omega)^n for all lambda > omega and all n. In words: if shifting A far to the right always gives a bounded inverse that shrinks at the rate 1/lambda (with the right power bounds), then A is a legitimate generator, and the semigroup it generates can be recovered from the resolvent.
Its importance is foundational: it is the theorem that makes the semigroup viewpoint usable. To prove a parabolic (or any linear) evolution equation is well-posed, you no longer chase the solution directly; you verify a resolvent estimate on the spatial operator — an elliptic a-priori bound you likely have anyway — and Hille-Yosida hands you existence, uniqueness, and continuous dependence in one stroke. The proof is constructive and gives the celebrated Yosida approximation: replace the unbounded A by bounded A_lambda = lambda A (lambda I - A)^{-1}, exponentiate those safely, and pass to the limit.
For A = Laplacian on L^2(Omega) with Dirichlet conditions, the elliptic estimate for lambda u - Laplacian u = f gives norm of u <= (1/lambda) norm of f for lambda > 0, i.e. norm of (lambda I - A)^{-1} <= 1/lambda. Hille-Yosida then immediately yields that A generates a contraction C_0 semigroup — so the heat IBVP is well-posed, no separate construction of e^{t Laplacian} required.
A resolvent bound on A is all you need to know the evolution is well-posed.
The 1/lambda bound is sharp and the power condition norm of (lambda I - A)^{-n} <= M/(lambda - omega)^n cannot be dropped to a single power when M > 1 — checking only one power is a classic mistake. For self-adjoint or analytic-semigroup cases there are friendlier equivalent forms (Lumer-Phillips, sectoriality), which is why Hille-Yosida in full generality is invoked less often than its specializations.