|
Analysis I by Herbert Amann, Page 373 函数列的极限的可导性
2.8 Theorem (differentiability of the limits of sequences of functions) Let $X$ be an open (or convex) perfect subset of $\mathbb{K}$ and $f_n \in C^1(X, E)$ for all $n \in \mathbb{N}$. Suppose that there are $f, g \in E^X$ such that
(i) $\left(f_n\right)$ converges pointwise to $f$, and
(ii) $\left(f_n^{\prime}\right)$ converges locally uniformly to $g$.
Then $f$ is in $C^1(X, E)$, and $f^{\prime}=g$. In addition, $\left(f_n\right)$ converges locally uniformly to $f$.
Proof. Let $a \in X$. Then there is some $r>0$ such that $\left(f_n^{\prime}\right)$ converges uniformly to $g$ on $B_r:=\mathbb{B}_{\mathbb{K}}(a, r) \cap X$. If $X$ is open we can choose $r>0$ so that $\mathbb{B}(a, r)$ is contained in $X$. Hence with either of our assumptions, $B_r$ is convex and perfect. Thus, for each $x \in B_r$, we can apply the mean value theorem (Theorem IV.2.18) to the function
$$
[0,1] \rightarrow E, \quad t \mapsto f_n(a+t(x-a))-t f_n^{\prime}(a)(x-a)
$$
to get
$$
\left|f_n(x)-f_n(a)-f_n^{\prime}(a)(x-a)\right| \leq \sup _{0<t<1}\left|f_n^{\prime}(a+t(x-a))-f_n^{\prime}(a)\right||x-a|
$$
Taking the limit $n \rightarrow \infty$ we get
$$\tag{2.1}\label1
|f(x)-f(a)-g(a)(x-a)| \leq \sup _{0<t<1}|g(a+t(x-a))-g(a)||x-a|
$$
for each $x \in B_r$. Theorem 2.4 shows that $g$ is in $C(X, E)$, so it follows from \eqref{1} that
$$
f(x)-f(a)-g(a)(x-a)=o(|x-a|) \quad(x \rightarrow a) .
$$
Hence $f$ is differentiable at $a$ and $f^{\prime}(a)=g(a)$. Since this holds for all $a \in X$, we have shown that $f \in C^1(X, E)$.
It remains to prove that $\left(f_n\right)$ converges locally uniformly to $f$. Applying the mean value theorem to the function
$$
[0,1] \rightarrow E, \quad t \mapsto\left(f_n-f\right)(a+t(x-a))
$$
we get the inequality
$$
\begin{aligned}
\left|f_n(x)-f(x)\right| & \leq\left|f_n(x)-f(x)-\left(f_n(a)-f(a)\right)\right|+\left|f_n(a)-f(a)\right| \\
& \leq r \sup _{0<t<1}\left|f_n^{\prime}(a+t(x-a))-f^{\prime}(a+t(x-a))\right|+\left|f_n(a)-f(a)\right| \\
& \leq r\left\|f_n^{\prime}-f^{\prime}\right\|_{\infty, B_r}+\left|f_n(a)-f(a)\right|
\end{aligned}
$$
for each $x \in B_r$. The right side of this inequality is independent of $x \in B_r$ and converges to 0 as $n \rightarrow \infty$ because of (ii) and the fact that $f^{\prime}=g$. Thus $\left(f_n\right)$ converges uniformly to $f$ on $B_r$. $_\blacksquare$
2.9 Corollary (differentiability of the limit of a series of functions) Suppose that $X \subseteq \mathbb{K}$ is open (or convex) and perfect, and $\left(f_n\right)$ is a sequence in $C^1(X, E)$ for which $\sum_{f_n} f_n$ converges pointwise and $\sum f_n^{\prime}$ converges locally uniformly. Then the $\operatorname{sum} \sum_{n=0}^{\infty} f_n$ is in $C^1(X, E)$ and
$$
\left(\sum_{n=0}^{\infty} f_n\right)^{\prime}=\sum_{n=0}^{\infty} f_n^{\prime} .
$$
In addition, $\sum f_n$ converges locally uniformly.
Proof This follows directly from Theorem 2.8. $_\blacksquare$
2.10 Remarks (a) Let $\left(f_n\right)$ be a sequence in $C^1(X, E)$ which converges uniformly to $f$. Even if $f$ is continuously differentiable, $\left(f_n^{\prime}\right)$ does not, in general, converge pointwise to $f^{\prime}$.
Proof Let $X:=\mathbb{R}, E:=\mathbb{R}$ and $f_n(x):=(1 / n) \sin (n x)$ for all $n \in \mathbb{N}^{\times}$. Because
$$
\left|f_n(x)\right|=|\sin (n x)| / n \leq 1 / n, \quad x \in X,
$$
$\left(f_n\right)$ converges uniformly to 0 . Since $\lim f_n^{\prime}(0)=1$, the sequence $\left(f_n^{\prime}(0)\right)$ does not converge to the derivative of the limit function at the point 0. $_\blacksquare$
(b) Let $\left(f_n\right)$ be a sequence in $C^1(X, E)$ such that $\sum f_n$ converges uniformly. Then, in general, $\sum f_n^{\prime}$ does not converge even pointwise.
Proof Suppose that $X:=\mathbb{R}, E:=\mathbb{R}$, and $f_n(x):=\left(1 / n^2\right) \sin (n x)$ for all $n \in \mathbb{N}^{\times}$. Then $\left\|f_n\right\|_{\infty}=1 / n^2$ and so, by the Weierstrass majorant criterion, the series $\sum f_n$ converges uniformly. Since $f_n^{\prime}(x)=(1 / n) \cos (n x), \sum f_n^{\prime}(0)$ does not converge. $_\blacksquare$ |
|