22.8. GENERAL THEORY OF CONTINUOUS SEMIGROUPS 605

≤∫

0

∥∥∥e−λ tS (t)∥∥∥∥∥∥∥S (h)x− x

h−Ax

∥∥∥∥dt

Then, passing to a limit as h→ 0+, this integrand converges uniformly to 0 so for allλ > α,

limh→0

R(λ )

(S (h)x− x

h

)= R(λ )Ax (22.32)

Also, S (h) commutes with R(λ ) . This follows from approximating with Riemann sumsand taking a limit. Thus also

limh→0

R(λ )

(S (h)x− x

h

)= lim

h→0

(S (h)R(λ )x−R(λ )x

h

)= AR(λ )x

so we have for x ∈ D(A) ,R(λ )Ax = AR(λ )x. However, this implies

R(λ )(λ I−A)x = (λ I−A)R(λ )x = x

from Claim 1. Thus R(λ ) is a left inverse of (λ I−A). Since R(λ ) = (λ I−A)−1 , thisshows the estimate 22.30 from 22.31. This proves Claim 3.

Why is A a closed operator? Suppose xn → x where xn ∈ D(A) and that Axn → ξ . Ineed to show that this implies that x ∈ D(A) and that Ax = ξ . Thus xn→ x and for λ > α,

(λ I−A)xn → λx− ξ . However, 22.30 shows that (λ I−A)−1 = R(λ ) is continuous andso

xn→ (λ I−A)−1 (λx−ξ ) = x

It follows that x ∈ D(A) . Then doing (λ I−A) to both sides of the equation, λx− ξ =λx−Ax and so Ax = ξ showing that A is a closed operator as claimed. ■

Definition 22.8.6 The linear mapping for λ > α where ∥S (t)∥ ≤ Meαt given by(λ I−A)−1 = R(λ ) is called the resolvent.

The following corollary is also very interesting.

Corollary 22.8.7 Let S (t) be a continuous semigroup and let A be its generator. Thenfor 0 < a < b < ∞ and x ∈D(A) , S (b)x−S (a)x =

∫ ba S (t)Axdt and also for t > 0 you can

take the derivative from the left,

limh→0+

S (t)x−S (t−h)xh

= S (t)Ax

Proof: Letting y∗ ∈ X ′, you can take y∗ inside the integral by approximating withRiemann sums. Thus

y∗(∫ b

aS (t)Axdt

)=∫ b

ay∗(

S (t) limh→0

S (h)x− xh

)dt

The difference quotients are bounded because they converge to Ax. Therefore, from thedominated convergence theorem and using the semigroup property,

y∗(∫ b

aS (t)Axdt

)= lim

h→0

∫ b

ay∗(

S (t)S (h)x− x

h

)dt

22.8. GENERAL THEORY OF CONTINUOUS SEMIGROUPS 605<| les [po _ Ax0 hThen, passing to a limit as h — 0+, this integrand converges uniformly to 0 so for allA>a,dtlim R (A) (Soe) = R(A)Ax (22.32)Also, S(h) commutes with R(A). This follows from approximating with Riemann sumsand taking a limit. Thus alsolim R (2) (see) tim (GOs e*) = AR(A)xso we have for x € D(A), R(A)Ax = AR(A) x. However, this impliesR(A) (AI—A)x = (AI—A)R(A)x =xfrom Claim 1. Thus R(A) is a left inverse of (AJ—A). Since R(A) = (AI—A)', thisshows the estimate 22.30 from 22.31. This proves Claim 3.Why is A a closed operator? Suppose x, — x where x, € D(A) and that Ax, > €. Ineed to show that this implies that x € D(A) and that Ax = €. Thus x, > x and forA > a,(AI —A) x, — Ax — €. However, 22.30 shows that (AJ —A)~! = R(A) is continuous andsoXp, > (AI—A) | (Ax—&) =xIt follows that x € D(A). Then doing (AJ—A) to both sides of the equation, Ax—§ =Ax —Ax and so Ax = & showing that A is a closed operator as claimed. HlDefinition 22.8.6 The linear mapping for A > a where \|S(t)|| < Me™ given by(AI—A)' =R(A) is called the resolvent.The following corollary is also very interesting.Corollary 22.8.7 Let S(t) be a continuous semigroup and let A be its generator. Thenfor0<a<b<eandx€ D(A), S(b)x—S(a)x= f? S(t) Axdt and also for t > 0 you cantake the derivative from the left,lim S()x—S(t-h)x = S(t) Axh-0+ hProof: Letting y* € X’, you can take y* inside the integral by approximating withRiemann sums. Thusy ([s@arar) = [y (sto)jim ==) dtThe difference quotients are bounded because they converge to Ax. Therefore, from thedominated convergence theorem and using the semigroup property,y* ([/s@anar) = jim [y' (sm) dt