Operator semigroups and evolution

A strongly continuous, or C0, semigroup on a Banach space consists of bounded linear maps T(t) for t≥0. They satisfy T(0)=I and T(t+s)=T(t)T(s), while T(t)x tends to x as t decreases to zero for every fixed x. The last condition is strong continuity. It does not require the operator norm of T(t)−I to tend to zero. General semigroups evolve forward only; an inverse for negative times may be unbounded or fail to exist.

The distinction between strong and norm continuity is visible in heat flow. For a Dirichlet eigenmode e_n on (0,π), T(t)e_n=e^{-n²t}e_n. Every fixed square-integrable profile approaches its initial value in L2 as t tends to zero. Yet the operator norm of T(t)−I is one for each positive t, because arbitrarily high modes can be almost completely damped immediately. One time step is small for each fixed state without being uniformly small across all unit states.

The generator A is defined by the limit of (T(h)x−x)/h as h decreases to zero. Its domain consists exactly of vectors for which that limit exists in the state-space norm. A C0 semigroup has a closed, densely defined generator, and T(t) preserves its domain. For x in D(A), the trajectory is differentiable and satisfies u′=Au. For general x in H, T(t)x still defines a continuous mild solution, but differentiability at the initial time is not automatic.

Not every closed densely defined operator generates a well-behaved forward semigroup. Additional estimates matter. A useful Hilbert-space case is a nonnegative self-adjoint operator B: the spectral calculus defines T(t)=exp(−tB), a contraction semigroup with generator −B. The sign suppresses high positive spectral values. Replacing that sign by plus generally produces unbounded forward amplification and can destroy bounded evolution on H. Generation theorems make these stability restrictions precise through resolvent bounds rather than formal notation.

For a continuous forcing f, the variation-of-constants expression combines the freely evolved initial state with the accumulated effect of past forcing. Under a C0 semigroup this gives a mild solution through a Banach-space integral. To differentiate it and call it a classical solution, one needs suitable domain and regularity assumptions. A formula can define a valid continuous state even when applying A pointwise to that state at every time has not been justified.

Heat flow smooths and loses high-frequency information, so its backward problem can amplify tiny errors enormously. A unitary group, by contrast, preserves norm and has a bounded inverse at every real time; self-adjoint quantum Hamiltonians generate such groups after multiplication by −i. Damping, unitary evolution, and transport are distinct structures. The finite sine-sum animation below merely changes common phases. It is not the heat semigroup and should not be used to infer a generator or a well-posed continuum evolution.