124 lines
4.7 KiB
TeX
124 lines
4.7 KiB
TeX
\section{Taylor's Formula}
|
|
\label{section:taylor}
|
|
|
|
\begin{theorem}[Taylor's Formula, Lagrange Remainder {{\cite[Theorem 4.7.1]{Bogachev}}}]
|
|
\label{theorem:taylor-lagrange}
|
|
Let $-\infty < a < b < \infty$, $E$ be a separated locally convex space, $S \subset [a, b]$ be at most countable, $n \in \natp$, and $f \in C^{n}([a, b]; E)$ be $(n+1)$-fold differentiable on $[a, b] \setminus N$, then
|
|
\[
|
|
f(b) - f(a) - \sum_{k = 1}^n \frac{1}{k!}D^kf(a)(b - a)^k
|
|
\]
|
|
|
|
is contained in the closed convex hull\footnote{It may be possible to sharpen the below claim to include the $1/(n+1)!$ factor. However, I was not able to follow the proof for this.} of
|
|
\[
|
|
\bracs{D^{n+1}f(s)(t - a)^{n+1} | s \in (a, b) \setminus N, t \in [a, b]}
|
|
\]
|
|
|
|
\end{theorem}
|
|
\begin{proof}
|
|
If $n = 0$, then the theorem is the \hyperref[Mean Value Theorem]{theorem:mean-value-theorem-line}.
|
|
|
|
Suppose inductively that the theorem holds for $n$. Let
|
|
\[
|
|
g: [a, b] \to E \quad t \mapsto f(t) - \sum_{k = 1}^{n+1} \frac{1}{k!}D^kf(a)(t - a)^k
|
|
\]
|
|
|
|
then for any $t \in (0, 1)$,
|
|
\begin{align*}
|
|
Dg(t) &= Df(t) - \sum_{k = 1}^{n+1} \frac{1}{(k-1)!}D^{k}f(a)(t - a)^{k-1} \\
|
|
&= Df(t) - Df(a) - \sum_{k = 1}^{n} \frac{1}{k!}D^{k+1}f(a)(t - a)^{k}
|
|
\end{align*}
|
|
by the \hyperref[Mean Value Theorem]{theorem:mean-value-theorem-line},
|
|
\[
|
|
g(b) - g(a) = f(b) - f(a) - \sum_{k = 1}^{n+1} \frac{1}{k!}D^kf(a)(t - a)^k
|
|
\]
|
|
|
|
is contained in the closed convex hull of
|
|
\[
|
|
\bracs{\braks{Df(t) - Df(a) - \sum_{k = 1}^{n} \frac{1}{k!}D^{k+1}f(a)(t - a)^{k}}(b - a) \bigg | t \in [a, b]}
|
|
\]
|
|
|
|
|
|
By the inductive hypothesis applied to $Df$, for any $t \in [a, b]$,
|
|
\[
|
|
Df(t) - Df(a) - \sum_{k = 1}^n \frac{1}{k!}D^{k+1}f(a)(t - a)^k
|
|
\]
|
|
|
|
is contained in the closed convex hull of
|
|
\[
|
|
\bracs{D^{n+2}f(s)(t - a)^{n+1} | s \in (a, t) \setminus N}
|
|
\]
|
|
|
|
Therefore
|
|
\[
|
|
f(b) - f(a) - \sum_{k = 1}^{n+1} \frac{1}{k!}D^kf(a)(b - a)^k
|
|
\]
|
|
|
|
is contained in the convex hull of
|
|
\[
|
|
\bracs{D^{n+2}f(s)(t - a)^{n+2} | s \in (a, t) \setminus N, t \in [a, b]}
|
|
\]
|
|
|
|
\end{proof}
|
|
|
|
\begin{theorem}[Taylor's Formula, Peano Remainder {{\cite[Theorem 4.7.3]{Bogachev}}}]
|
|
\label{theorem:taylor-peano}
|
|
Let $E$ be a topological vector space, $\sigma \subset B(E)$ be an upward-directed family that includes bounded sets contained in finite-dimensional subspaces, $F$ be a separated locally convex space, $U \subset E$ be open, and $f: U \to F$ be $n$-fold $\sigma$-differentiable at $x_0 \in U$, then there exists $r \in \mathcal{R}_\sigma^n(E; F)$ such that
|
|
\[
|
|
g(x_0 + h) = g(x_0) + \sum_{k = 1}^n \frac{1}{k!}D^k_\sigma f(x_0)(h^{(k)}) + r(h)
|
|
\]
|
|
|
|
\end{theorem}
|
|
\begin{proof}
|
|
Let
|
|
\[
|
|
r(h) = g(x_0 + h) - g(x) - \sum_{k = 1}^n \frac{1}{k!}D^k_\sigma f(x_0)(h^{(k)})
|
|
\]
|
|
|
|
For any $1 \le k \le n$, $D^k_\sigma(x_0) \in B^k_\sigma(E; F)$ is symmetric by \autoref{theorem:derivative-symmetric}. Let $T_k(h) = \frac{1}{k!}D^k_\sigma f(x_0)(h^{(k)})$, then by \autoref{proposition:multilinear-derivative}, for any $\bracs{t_j}_1^\ell \in E$,
|
|
\[
|
|
D^\ell_\sigma T_k(h)(t_1, \cdots, t_\ell) = \begin{cases}
|
|
0 &\ell > k \\
|
|
D^k_\sigma(x_0)(t_1, \cdots, t_\ell) &\ell = k \\
|
|
\frac{1}{(k-\ell)!}D^k_\sigma(x_0)(h^{(k - \ell)}, t_1, \cdots, t_\ell) & \ell < k
|
|
\end{cases}
|
|
\]
|
|
|
|
so
|
|
\[
|
|
D^k_\sigma r(0) = D^k_\sigma g(x_0) - D^k_\sigma(x_0) = 0
|
|
\]
|
|
|
|
|
|
If $n = 1$, then the theorem holds by definition of the derivative. Now suppose inductively that the theorem holds for $n$. By the \hyperref[Mean Value Theorem]{theorem:mean-value-theorem},
|
|
\[
|
|
r(h) = r(h) - r(0) \in \overline{\text{Conv}\bracs{D_\sigma r(s)(h)| s \in [0, h]}}
|
|
\]
|
|
|
|
For any $A \in \sigma$ and $t > 0$,
|
|
\[
|
|
\frac{r(tA)}{t^{n+1}} \subset \overline{\text{Conv}\bracs{t^{-n}D_\sigma r(s)(h)|s \in [0, h], h \in tA}}
|
|
\]
|
|
|
|
Let $U \in \cn_F(0)$ be convex and circled, then by the inductive assumption applied to $D_\sigma r$, there exists $t_0 \in (0, 1)$ such that for any $t \in (0, t_0)$.
|
|
\[
|
|
\frac{D_\sigma r(tA)}{t^n} \subset \bracs{T \in L(E; F)| T(A) \subset U}
|
|
\]
|
|
|
|
Since $U$ is circled,
|
|
\[
|
|
\bracs{T \in L(E; F)| T(A) \subset U} = \bracs{T \in L(E; F)| T(tA) \subset U \forall t \in (0, 1)}
|
|
\]
|
|
|
|
so
|
|
\[
|
|
\bracs{t^{-n}D_\sigma r(s)(h)|s \in [0, h], h \in tA} \subset U
|
|
\]
|
|
|
|
and
|
|
\[
|
|
\frac{r(tA)}{t^{n+1}} \subset \overline{\text{Conv}\bracs{t^{-n}D_\sigma r(s)(h)|s \in [0, h], h \in tA}} \subset \overline{U}
|
|
\]
|
|
|
|
Therefore $r \in \mathcal{R}_\sigma^{n+1}(E; F)$.
|
|
\end{proof}
|