C0-Semigroups and the Hille-Yosida Theorem

Intuition Beginner

Imagine a system that evolves in time according to a fixed rule: a metal bar cooling, a population diffusing, a quantum particle drifting. If you know the state now, the rule hands you the state one second later, and two seconds later, and so on. Run the rule for $t$ seconds, then run it again for $s$ seconds, and you have run it for $t+s$ seconds total. This composition law is the whole idea of a semigroup: a family of "time-advance machines" $T(t)$, one for each duration $t\ge 0$, that stack by adding their times.

The cleanest example is the ordinary exponential. The scalar equation $u^{\prime }=au$ has solution $u(t)=e^{ta}u(0)$, and $e^{ta}{e}^{sa}=e^{(t+s)a}$. The number $a$ is the "rule"; the exponential turns it into a flow. For a vector of unknowns the rule becomes a matrix $A$ and the flow becomes the matrix exponential $e^{tA}$.

The leap is to let the state live in an infinite-dimensional space of functions, and let $A$ be a differential operator like the Laplacian. Now there is no convergent power series for $e^{tA}$, but the flow $T(t)$ still exists. The question this unit answers is: which rules $A$ generate a sensible flow?

Visual Beginner

A semigroup is a one-way clock of operators. At time zero it does nothing; as time advances it pushes every starting state along its trajectory, and the pushes compose by adding times.

The generator $A$ is the velocity of this flow at the starting line: it records how fast a state begins to move the instant the clock starts. Knowing the generator is knowing the rule; the flow is its time-integrated form.

Worked example Beginner

Take the simplest infinite-dimensional flow: rigid translation. Let a state be a function $f(x)$ on the line, and let the rule be "shift to the right at unit speed". After time $t$ the machine returns $T(t)f$, the function whose value at $x$ is $f(x-t)$.

Step 1. Check the start. At $t=0$ the shift is by zero, so $T(0)f=f$. The machine does nothing at time zero.

Step 2. Check composition. Shift by $t$, then by $s$. First $T(t)f$ has value $f(x-t)$ at $x$. Applying $T(s)$ shifts again: the result at $x$ is the previous function evaluated at $x-s$, namely $f((x-s)-t)=f(x-t-s)$. That is exactly $T(t+s)f$. So $T(t)T(s)=T(t+s)$.

Step 3. Find the velocity. Take $f(x)=x^2$ and watch the value at a fixed point $x=3$. We have $T(t)f$ at $x=3$ equal to $(3-t{)}^2=9-6t+t^2$. Its rate of change at $t=0$ is $-6$. The general rule: the starting velocity is minus the slope of $f$. The generator of right-translation is "minus the derivative".

What this tells us: a concrete flow (sliding a graph) hides a generator (a derivative operator). The flow is smooth in time even though the generator involves differentiation, the operation that usually roughens functions. The semigroup tames the operator.

Check your understanding Beginner

Formal definition Intermediate+

Let $X$ be a Banach space 02.11.04. A strongly continuous semigroup (or $C_0$-semigroup) is a family $\{T(t){\}}_{t\ge 0}$ of bounded linear operators $T(t)\in \mathcal{B}(X)$ satisfying

$$T(0)=I,T(t+s)=T(t)T(s)(t,s\ge 0),$$

together with the strong continuity axiom

$$\underset{t\to 0^{+}}{\lim }\Vert T(t)x-x\Vert =0\text{for every }x\in X.$$

Strong continuity is weaker than norm continuity of $t\mapsto T(t)$; the heat semigroup is strongly but not norm continuous at $t=0$.

A standard consequence is an exponential bound: there exist $M\ge 1$ and $\omega \in \mathbb{R}$ with

$$\Vert T(t)\Vert \le M{e}^{\omega t}(t\ge 0)[\text{ref: TODO\_REF Pazy 1983, §1.2}].$$

When $M=1$ and $\omega =0$ the semigroup is a contraction semigroup: $\Vert T(t)\Vert \le 1$.

The infinitesimal generator of the semigroup is the operator $A$ defined by

$$Ax=\underset{t\to 0^{+}}{\lim }\frac{T(t)x-x}{t},$$

with domain $D(A)$ equal to the set of $x\in X$ for which this limit exists in the norm of $X$. The generator is in general an unbounded operator 02.11.03: $D(A)$ is a proper dense subspace and $A$ is closed.

For $\lambda \in \mathbb{C}$, the resolvent set $\rho (A)$ consists of those $\lambda$ for which $\lambda I-A:D(A)\to X$ is a bijection with bounded inverse, the resolvent

$$R(\lambda ,A)=(\lambda I-A{)}^{-1}\in \mathcal{B}(X).$$

An operator $A$ is dissipative if for every $x\in D(A)$ there is a normalised duality pairing witnessing $\mathrm{Re}⟨Ax,x^{\ast }⟩\le 0$; concretely, on a Hilbert space $A$ is dissipative iff $\mathrm{Re}⟨Ax,x⟩\le 0$ for all $x\in D(A)$.

Counterexamples to common slips Intermediate+

• Strong continuity is not norm continuity. If $t\mapsto T(t)$ were norm continuous at $0$, then $A$ would be bounded and $T(t)=e^{tA}$ a convergent series. Genuine PDE semigroups (heat, wave) have unbounded generators precisely because they are only strongly continuous.

• The generator is not defined everywhere. Writing "$A=$ the derivative" hides the domain. The translation semigroup on $L^2(\mathbb{R})$ has generator $-\frac{d}{dx}$ on the domain $H^1(\mathbb{R})$, not on all of $L^2$. The same formula on a different domain is a different operator and may fail to generate a semigroup.

• Dissipativity alone does not generate. A dissipative operator generates a contraction semigroup only when a range condition also holds: $\mathrm{ran}(\lambda I-A)=X$ for some (hence all) $\lambda >0$. Dropping the range condition leaves merely symmetric-looking operators that have no flow.

Key theorem with proof Intermediate+

Theorem (Hille-Yosida, contraction case). A linear operator $A$ on a Banach space $X$ is the infinitesimal generator of a $C_0$-semigroup of contractions $\{T(t){\}}_{t\ge 0}$ if and only if

1. $A$ is closed and densely defined, and
2. every $\lambda >0$ lies in $\rho (A)$ and the resolvent satisfies

$$\Vert R(\lambda ,A)\Vert \le \frac{1}{\lambda }(\lambda >0)[\text{ref: TODO\_REF Yosida 1948}].$$

Proof. ($\Rightarrow$) Suppose $A$ generates a contraction semigroup. Density of $D(A)$ and closedness of $A$ are standard properties of generators (proved in the Full proof set). For $\lambda >0$ define

$$R(\lambda )x={\int }_0^{\infty }{e}^{-\lambda t}T(t)xdt,$$

a convergent Bochner integral since $\Vert e^{-\lambda t}T(t)x\Vert \le e^{-\lambda t}\Vert x\Vert$. A direct manipulation using the semigroup law shows $R(\lambda )=(\lambda I-A{)}^{-1}$: for $x\in D(A)$, integrating by parts in $t$ against $\frac{d}{dt}T(t)x=T(t)Ax$ gives $R(\lambda )(\lambda I-A)x=x$, and the reverse identity follows similarly. The contraction bound yields

$$\Vert R(\lambda )x\Vert \le {\int }_0^{\infty }{e}^{-\lambda t}\Vert T(t)x\Vert dt\le \Vert x\Vert {\int }_0^{\infty }{e}^{-\lambda t}dt=\frac{\Vert x\Vert }{\lambda }.$$

($\Leftarrow$) Assume (1) and (2). The construction is the Yosida approximation. For $\lambda >0$ set

$$A_{\lambda }=\lambda AR(\lambda ,A)={\lambda }^{2}R(\lambda ,A)-\lambda I\in \mathcal{B}(X),$$

the second equality from $\lambda AR(\lambda ,A)=\lambda (\lambda I-(\lambda I-A))R(\lambda ,A)={\lambda }^{2}R(\lambda ,A)-\lambda I$. Each $A_{\lambda }$ is bounded, so the exponentials $e^{t{A}_{\lambda }}$ are defined by their power series and form uniformly continuous semigroups.

First, $\lambda R(\lambda ,A)x\to x$ as $\lambda \to \infty$ for every $x\in X$. For $x\in D(A)$ one has $\lambda R(\lambda ,A)x-x=R(\lambda ,A)Ax$, whose norm is $\le \Vert Ax\Vert /\lambda \to 0$; density of $D(A)$ and the uniform bound $\Vert \lambda R(\lambda ,A)\Vert \le 1$ extend the limit to all of $X$. Consequently $A_{\lambda }x=\lambda R(\lambda ,A)Ax\to Ax$ for $x\in D(A)$.

Next, each $e^{t{A}_{\lambda }}$ is a contraction. Writing $A_{\lambda }={\lambda }^{2}R(\lambda ,A)-\lambda I$,

$$\Vert e^{t{A}_{\lambda }}\Vert =e^{-\lambda t}\Vert e^{t{\lambda }^{2}R(\lambda ,A)}\Vert \le e^{-\lambda t}{e}^{t{\lambda }^2\Vert R(\lambda ,A)\Vert }\le e^{-\lambda t}{e}^{\lambda t}=1,$$

using $\Vert R(\lambda ,A)\Vert \le 1/\lambda$. The bounded generators $A_{\lambda }$ and $A_{\mu }$ commute (both are functions of the commuting resolvents), so for $x\in D(A)$,

$$\Vert e^{t{A}_{\lambda }}x-e^{t{A}_{\mu }}x\Vert =\Vert {\int }_0^t\frac{d}{ds}\left(e^{(t-s)A_{\mu }}{e}^{s{A}_{\lambda }}x\right)ds\Vert \le {\int }_0^t\Vert e^{(t-s)A_{\mu }}{e}^{s{A}_{\lambda }}(A_{\lambda }-A_{\mu })x\Vert ds\le t\Vert A_{\lambda }x-A_{\mu }x\Vert .$$

The right side tends to $0$ as $\lambda ,\mu \to \infty$, uniformly for $t$ in compact intervals. Thus $T(t)x={\lim }_{\lambda \to \infty }{e}^{t{A}_{\lambda }}x$ exists for $x\in D(A)$, uniformly on compacta, and extends by density and the uniform contraction bound to all $x\in X$. The limit family $\{T(t)\}$ inherits $T(0)=I$, the semigroup law, contractivity, and strong continuity from the approximants. A final computation identifies the generator of $\{T(t)\}$ with $A$: for $x\in D(A)$, $T(t)x-x={\lim }_{\lambda }{\int }_0^t{e}^{s{A}_{\lambda }}{A}_{\lambda }xds={\int }_0^{t}T(s)Axds$, whose derivative at $t=0$ is $Ax$. $\square$

Bridge. The Yosida approximation is exactly the device that builds toward every later existence theorem for evolution equations: it replaces an unbounded generator by bounded surrogates $A_{\lambda }$ whose exponentials are honest power series, then passes to the limit. This is the foundational reason the abstract Cauchy problem $u^{\prime }=Au$ is well posed precisely when $A$ generates a semigroup, and it generalises the matrix exponential $e^{tA}$ of finite-dimensional linear ODE theory to operators with no convergent series. The resolvent bound $\Vert R(\lambda ,A)\Vert \le 1/\lambda$ is dual to the contraction bound $\Vert T(t)\Vert \le 1$ through the Laplace transform $R(\lambda ,A)={\int }_0^{\infty }{e}^{-\lambda t}T(t)dt$; this is exactly the analytic content the theorem extracts. The same circle of ideas appears again in 02.13.03 the heat semigroup $e^{t\Delta }$, where dissipativity of the Laplacian is the input, and the central insight that a flow is recovered from the velocity at the start carries into the analytic-semigroup theory of parabolic problems. Putting these together, Hille-Yosida is the gateway from static functional analysis to dynamics.

Exercises Intermediate+

Lean formalization Intermediate+

lean_status: none — current Mathlib does not package the unbounded-generator $C_0$-semigroup theory used here.

The expected formalization would introduce a C0Semigroup X structure bundling the family T : ℝ≥0 → (X →L[𝕜] X) with the identity, semigroup-law, and strong-continuity fields; define the generator as a densely defined operator on a submodule domain via the strong limit of difference quotients; build the resolvent R(λ, A) = (λ • 1 - A)⁻¹ and the Laplace representation R(λ, A) x = ∫ t in Ioi 0, exp (-λ * t) • T t x; and state Hille-Yosida as the equivalence between generation and the resolvent bound ‖(R(λ, A))^[n]‖ ≤ M / (λ - ω)^n, with the Yosida approximation A_λ = λ • A ∘ R(λ, A) as the constructive witness.

Advanced results Master

General Hille-Yosida and the power bound. Dropping contractivity, $A$ generates a $C_0$-semigroup with $\Vert T(t)\Vert \le M{e}^{\omega t}$ if and only if $A$ is closed, densely defined, $(\omega ,\infty )\subset \rho (A)$, and

$$\Vert R(\lambda ,A{)}^n\Vert \le \frac{M}{(\lambda -\omega {)}^n}(\lambda >\omega ,n\ge 1).$$

The power on the resolvent is essential: a bound on $\Vert R(\lambda ,A)\Vert$ alone controls $M=1$ but not general $M$, because $R(\lambda ,A{)}^n=\frac{(-1{)}^{n-1}}{(n-1)!}\frac{d^{n-1}}{d{\lambda }^{n-1}}R(\lambda ,A)$ encodes the $n$-fold iterated Laplace transform ${\int }_0^{\infty }\frac{t^{n-1}}{(n-1)!}{e}^{-\lambda t}T(t)dt$.

Lumer-Phillips theorem. A densely defined operator $A$ generates a $C_0$-semigroup of contractions if and only if $A$ is dissipative and $\mathrm{ran}({\lambda }_{0}I-A)=X$ for some ${\lambda }_0>0$. Dissipativity gives $\Vert (\lambda I-A)x\Vert \ge \lambda \Vert x\Vert$, hence injectivity and the resolvent bound; the range condition supplies surjectivity. This is the working criterion in PDE, because dissipativity is an energy estimate and the range condition is solvability of $\lambda u-Au=f$ ^{[Lumer-Phillips 1961]}.

Stone's theorem. On a Hilbert space, $A$ generates a $C_0$-group of unitaries $\{T(t){\}}_{t\in \mathbb{R}}$ if and only if $A=iH$ with $H$ self-adjoint 02.11.03. Then $T(t)=e^{itH}$, the time evolution of quantum mechanics; the Schrödinger group and the wave group are the canonical instances ^{[Stone 1932]}.

Analytic semigroups and sectoriality. An operator $A$ is sectorial of angle $\theta \in (0,\pi /2]$ if its resolvent set contains a sector ${\Sigma }_{\pi /2+\theta }=\{\lambda \ne 0:|\arg \lambda |<\pi /2+\theta \}$ and $\Vert R(\lambda ,A)\Vert \le M/|\lambda |$ there. Equivalently, $A$ generates a semigroup $T(t)$ that extends holomorphically to a sector in the complex $t$-plane and satisfies the smoothing estimate

$$\Vert AT(t)\Vert \le \frac{C}{t}(t>0).$$

The estimate $\Vert AT(t)\Vert \le C/t$ is the abstract form of parabolic smoothing: it says the semigroup instantly maps $X$ into $D(A)$, the analytic content of "the heat equation smooths". The Laplacian on $L^p({\mathbb{R}}^n)$ is sectorial of angle $\pi /2$, so $e^{t\Delta }$ is analytic 02.13.03.

Synthesis. The four generation theorems are one structure seen from four angles, and putting these together exposes the central insight of the subject: a linear evolution is determined by the velocity field $A$ at the origin of time, and the analytic fingerprint of that field is its resolvent. Hille-Yosida reads generation off the iterated resolvent bound; Lumer-Phillips reads it off dissipativity, which is exactly the energy estimate a PDE supplies, so the abstract theory and the concrete a priori estimate are dual descriptions of the same flow. Stone's theorem is the foundational reason quantum dynamics is unitary: self-adjointness of the Hamiltonian is the generation condition for a group rather than a semigroup, and the reversibility of $e^{itH}$ generalises the finite-dimensional fact that skew-Hermitian matrices exponentiate to unitaries. Analytic-semigroup sectoriality, with its smoothing bound $\Vert AT(t)\Vert \le C/t$, is dual to dissipativity-plus-resolvent-decay and is exactly the abstract statement that parabolic flows regularise instantly; this is the bridge from the heat semigroup $e^{t\Delta }$ of 02.13.03 to the full nonlinear theory, where the generation theory of the linearisation appears again as the engine of fixed-point existence proofs for quasilinear and reaction-diffusion equations.

Full proof set Master

Proposition 1 (the generator is densely defined and closed). Let $\{T(t)\}$ be a $C_0$-semigroup with generator $A$. Then $D(A)$ is dense in $X$ and $A$ is a closed operator.

Proof. For $x\in X$ and $r>0$ set $x_r=\frac{1}{r}{\int }_0^{r}T(s)xds$, a convergent Riemann/Bochner integral by strong continuity. Then

$$\frac{T(h)-I}{h}{x}_r=\frac{1}{h}\left(\frac{1}{r}{\int }_0^{r}T(s+h)xds-\frac{1}{r}{\int }_0^{r}T(s)xds\right)=\frac{1}{r}\cdot \frac{1}{h}\left({\int }_r^{r+h}T(s)xds-{\int }_0^{h}T(s)xds\right).$$

As $h\to 0^{+}$ the right side tends to $\frac{1}{r}(T(r)x-x)$ by continuity of $s\mapsto T(s)x$. Hence $x_r\in D(A)$ with $A{x}_r=\frac{1}{r}(T(r)x-x)$. Moreover $x_r\to x$ as $r\to 0^{+}$, again by strong continuity, so $D(A)$ is dense.

Closedness was established in Exercise 6 via $T(t)x_n-x_n={\int }_0^{t}T(s)A{x}_{n}ds$: if $x_n\to x$ and $A{x}_n\to y$, the limit identity $T(t)x-x={\int }_0^{t}T(s)yds$ divided by $t$ and sent to $0$ gives $x\in D(A)$, $Ax=y$. $\square$

Proposition 2 (dissipativity yields the resolvent bound). If $A$ is dissipative, then for every $\lambda >0$ and $x\in D(A)$,

$$\Vert (\lambda I-A)x\Vert \ge \lambda \Vert x\Vert .$$

In particular $\lambda I-A$ is injective with $\Vert R(\lambda ,A)\Vert \le 1/\lambda$ whenever it is surjective.

Proof. By dissipativity choose a norming functional $x^{\ast }\in X^{\ast }$ with $\Vert x^{\ast }\Vert =1$, $⟨x,x^{\ast }⟩=\Vert x\Vert$, and $\mathrm{Re}⟨Ax,x^{\ast }⟩\le 0$. Then

$$\Vert (\lambda I-A)x\Vert \ge \mathrm{Re}⟨(\lambda I-A)x,x^{\ast }⟩=\lambda \Vert x\Vert -\mathrm{Re}⟨Ax,x^{\ast }⟩\ge \lambda \Vert x\Vert .$$

Injectivity is immediate. If $\lambda I-A$ is also onto, then for $y=(\lambda I-A)x$ we get $\Vert R(\lambda ,A)y\Vert =\Vert x\Vert \le \Vert y\Vert /\lambda$, i.e. $\Vert R(\lambda ,A)\Vert \le 1/\lambda$. $\square$

Proposition 3 (uniqueness of the semigroup from its generator). A $C_0$-semigroup is determined by its generator: if two $C_0$-semigroups $\{T(t)\}$ and $\{S(t)\}$ have the same generator $A$, then $T(t)=S(t)$ for all $t\ge 0$.

Proof. Fix $t>0$ and $x\in D(A)$. The map $s\mapsto S(t-s)T(s)x$ on $[0,t]$ is differentiable with derivative $-AS(t-s)T(s)x+S(t-s)AT(s)x=0$, using that $S$ and $T$ have generator $A$ and that $T(s)x\in D(A)$. Hence the map is constant; comparing $s=0$ and $s=t$ gives $S(t)x=T(t)x$. Density of $D(A)$ and boundedness of $S(t),T(t)$ extend the equality to all of $X$. $\square$

Connections Master

• Unbounded self-adjoint operators 02.11.03 — Stone's theorem identifies generators of unitary groups with $iH$, $H$ self-adjoint; the spectral theorem there builds $e^{itH}$ and the deficiency-index theory decides which symmetric operators generate a group at all.

• Banach spaces 02.11.04 — the entire theory lives on a Banach space; the resolvent bound, the Bochner integral ${\int }_0^{\infty }{e}^{-\lambda t}T(t)dt$, and the uniform boundedness of $\{e^{t{A}_{\lambda }}\}$ are Banach-space estimates, and contractivity is measured in the Banach norm.

• Heat equation, heat kernel, and Duhamel's principle 02.13.03 — the heat semigroup $e^{t\Delta }f=\Phi (\cdot ,t)\ast f$ is the prototype analytic semigroup; its generator is the Laplacian, dissipativity is the energy inequality, and Duhamel's principle is the variation-of-parameters formula $u(t)=T(t)u_0+{\int }_0^{t}T(t-s)f(s)ds$ for the inhomogeneous Cauchy problem.

• Poisson equation and fundamental solution 02.13.02 — solving $\lambda u-\Delta u=f$ for the resolvent of the Laplacian is the elliptic range condition feeding Lumer-Phillips, linking semigroup generation to elliptic solvability.

Historical & philosophical context Master

The theory crystallised in 1948 when Einar Hille and Kosaku Yosida independently characterised the generators of strongly continuous contraction semigroups; Yosida's note in the Journal of the Mathematical Society of Japan introduced the approximation $A_{\lambda }=\lambda AR(\lambda ,A)$ that bears his name ^{[Yosida 1948]}, and Hille's account appeared in the 1948 first edition of his Colloquium volume, expanded with Ralph Phillips in the 1957 Functional Analysis and Semi-Groups ^{[Hille 1957]}. Marshall Stone had already proved the unitary-group case in 1932 ^{[Stone 1932]}, giving the mathematical foundation for the time evolution in von Neumann's formulation of quantum mechanics. Günter Lumer and Ralph Phillips recast generation through dissipativity in 1961 ^{[Lumer-Phillips 1961]}, producing the criterion that PDE practitioners use because dissipativity is an energy estimate. Analytic-semigroup theory grew from the parabolic regularity results of the 1950s and 1960s and reached its modern functional-analytic form in the monographs of Pazy and of Engel and Nagel.

Bibliography Master

• Hille, E. and Phillips, R. S., Functional Analysis and Semi-Groups, American Mathematical Society Colloquium Publications 31, 1957.
• Yosida, K., "On the differentiability and the representation of one-parameter semi-groups of linear operators", Journal of the Mathematical Society of Japan 1 (1948), 15-21.
• Stone, M. H., "On one-parameter unitary groups in Hilbert space", Annals of Mathematics 33 (1932), 643-648.
• Lumer, G. and Phillips, R. S., "Dissipative operators in a Banach space", Pacific Journal of Mathematics 11 (1961), 679-698.
• Pazy, A., Semigroups of Linear Operators and Applications to Partial Differential Equations, Springer Applied Mathematical Sciences 44, 1983.
• Engel, K.-J. and Nagel, R., One-Parameter Semigroups for Linear Evolution Equations, Springer Graduate Texts in Mathematics 194, 2000.

---

Analysis-side parabolic-hyperbolic chapter unit. The abstract generation theory underlying modern evolution-equation existence.