Question about a step in the proof of Thm. 1.2 of Pazy (Semigroups), Section 4.1

63 Views Asked by At

For anyone who happens to have Pazy's book on hand, on page 102, we have $$\frac{d}{dt}R(\lambda,A)u(t)=\lambda R(\lambda,A)u(t)-u(t)$$ which implies $$R(\lambda,A)u(t)=-\int_{0}^{t}e^{\lambda(t-\tau)}u(\tau)d\tau$$ I suppose I am probably missing something obvious, but why is this implication valid?

For some background, $A$ is a densely-defined operator with $[0,\infty)\subset\rho(A)$, and $$\limsup_{\lambda\to\infty}\lambda^{-1}\log||R(\lambda,A)||\leq 0$$ We are trying to show that for any initial data in the Banach space, $X$, on which $A$ is defined, there can be at most one solution to the Abstract Cauchy Problem for $A$. Accordingly, $u$ here is assumed to solve this problem for initial data $\vec{0}$. Pazy's proof also depends on Lemma 1.1 of this section and a translation argument as well, both of which make sense to me. I am just stuck on the one step above.

1

There are 1 best solutions below

1
On BEST ANSWER

This is merely the variation of constant formula :

For any simple Cauchy problem : $v'(t)=Bv(t)+f(t); \; v(0)=x\in X$, where $B$ generates a $C_0$-semigroup $T(t)$ on a Banach space $X$, the solution (when it exists) is given by the classical formula : $$v(t)=T(t)x+\int_0^t T(t-s) f(s)ds.$$ Now, in your case, we have $v(t)=R(\lambda,A)u(t), \; x=v(0)=0$ (since $u(0)=0$), $B=\lambda I$, $T(t)=e^{t \lambda I}=e^{\lambda t}I$ and $ f(t)=-u(t)$.

We can also prove this result as in the proof of variation constant formula. In fact, take $v(\tau)=R(\lambda,A)u(\tau)$ then your fisrt formula is equivalent to $$v'(\tau)-\lambda v(\tau)=-u(\tau).$$ Multiplying by $e^{-\lambda \tau}$, we obtain $$e^{-\lambda \tau}v'(\tau)-\lambda e^{-\lambda \tau}v(\tau)=-e^{-\lambda \tau}u(\tau).$$ Then $$\frac{d}{dt} [e^{-\lambda \tau}v(\tau)]=-e^{-\lambda \tau}u(\tau).$$ Integrating from $0$ to $t$ and using $v(0)=0$, we have $$e^{-\lambda t}v(t)=-\int_0^t e^{-\lambda \tau}u(\tau) d\tau.$$ Thus, $$v(t)=R(\lambda,A)u(t)=-\int_{0}^{t}e^{\lambda(t-\tau)}u(\tau)d\tau.$$