Skip to main content

Section 9.4 Solutions to Abstract ODEs

We consider the abstract ODE on a Banach space \(X\) with unbounded operator \(A\) as
\begin{equation} u_t = Au + f(t), \qquad u(0)= u_0 \in X,\label{abstractODE}\tag{9.4.1} \end{equation}
where we assume \(u_0\in D(A)\text{,}\) \(f\in C([0,T], X) \text{.}\) For \(T>0\) we are interested in a classical solution
\begin{equation*} u\in C^1([0,T], X) \cap C([0,T], D(A)). \end{equation*}
This is, in fact, the classical variation of constant formula. Using the properties of the generator in Theorem (Theorem 9.3.3), we directly compute for \(g(s) = T(t-s) u(s)\) that
\begin{align*} \frac{dg}{ds} \amp=\amp - A T(t-s) u(s) + T(t-s) \dot u(s)\\ \amp=\amp - A T(t-s) u(s) +T(t-s)(A u(s) + f(s))\\ \amp=\amp T(t-s) f(s). \end{align*}
Hence
\begin{equation*} g(t) - g(0)= u(t) - T(t) u_0 = \int_0^t T(t-s) f(s) ds . \end{equation*}
Definition 9.4.2.
Given
\begin{equation*} \dot u = Au + f(t), \qquad \mbox{in }\quad X, \end{equation*}
and \(f\in L^1((0,T), X)\text{.}\) Then \(u\) is called a mild solution if it satisfies the variation of constant formula (Theorem 9.4.1).
Remark 9.4.3.
Note that a mild solution does not need any differentiation.
Remark 9.4.4.
Note that a classical solution is automatically a mild solution. However, a mild solution is not necessarily a weak or classical solution. But mild solutions can be used to find those.
The next result answers the question when a mild solution is classical.
By Theorem (Theorem 9.3.3) the term \(T(t) u_0\) is differentiable and \(T(t)u_0\in D(A)\text{,}\) hence the first term on the right hand side of (Theorem 9.4.1) is classical. Now we study the integral term
\begin{equation} v(t) := \int_0^t T(t-s) f(s) ds .\label{defv}\tag{9.4.3} \end{equation}
We need to show that
\begin{equation*} v\in \underbrace{C^1([0,T], X)}_{(1)}\cap \underbrace{C([0,T], D(A))}_{(2)}. \end{equation*}
We compute
\begin{align*} \underbrace{\frac{T(h) - I}{h}}_{(3)} v \amp=\amp \frac{1}{h} \left( \int_0^t T(h+t-s) f(s) ds - \int_0^t T(t-s) f(s) ds \right)\\ \amp=\amp \frac{1}{h} \left( \underbrace{\int_0^{t+h} T(h+t-s) f(s) ds}_{v(t+h)} - \int_t^{t+h} T(h+t-s) f(s) ds - \underbrace{\int_0^t T(t-s) f(s) ds}_{v(t)} \right)\\ \amp=\amp \underbrace{\frac{1}{h} (v(t+h)-v(t))}_{(4)} - \frac{1}{h} \int_t^{t+h} T(h+t-s) f(s) ds. \end{align*}
From Theorem (Theorem 9.3.3) we evaluate the limit of the last term
\begin{equation*} \lim_{h\to 0} \frac{1}{h} \int_t^{t+h} T(h+t-s) f(s) ds = f(t). \end{equation*}
The limits of (3) and (4) need some more attention. In fact, if (1) is true, then (4) converges
\begin{equation*} \lim_{h\to 0} \frac{1}{h}(v(t+h)-v(t)) = \dot v(t). \end{equation*}
In this case (3) converges to \(Av\) and (2) is true, i.e (1) implies (2).

If (2) is true then (3) converges
\begin{equation*} \lim_{h\to 0} \frac{T(h) - I}{h} v(t) = A v(t) . \end{equation*}
Moreover, if (3) converges, then also (4), which implies (1). Hence (1) implies (2). Also, if (4) converges, then also (3). We find that (1) implies (2) and vice versa. We break this circle conclusion by the additional conditions on \(f\text{.}\)

If \(f\in L^1(0,T, D(A))\text{,}\) then (3) converges by the definition of \(v(t)\) in ((9.4.3)). If \(f\in W^{1,1}([0,T], X)\) we use the convolution identity
\begin{equation*} v(t) = \int_0^t T(s) f(t-s) ds \end{equation*}
and differentiate
\begin{equation*} \dot v(t) = T(t) f(0) + \int_0^t T(s) \dot f(t-s) ds , \end{equation*}
which is well defined. Hence (1) holds. Altogether we have a classical solution of \(\dot u = Au + f(t)\text{,}\) \(u(0)=u_0\text{.}\)