Theorem 2.1.3.label Let $E, F$ be Banach spaces, $U \subset E$ and $V \subset F$ be open subsets, and $f \in C^{k}(U \times V; L(E; F))$ ($k \ge 2$).

For any $\xi, \eta \in C^{k}(U \times V; E)$, denote

\[\Xi(x, y) = (\xi(x, y), f(x, y) \cdot \xi(x, y)) \quad H(x, y) = (\eta(x, y), f(x, y) \cdot \eta(x, y))\]

If for every $\xi, \eta \in C^{k}(U \times V; E)$ and $(x, y) \in U \times V$,

\[[Df(x, y) \cdot \Xi(x, y)] \cdot \eta(x, y) = [Df(x, y) \cdot H(x, y)] \cdot \xi(x, y)\]

then for every $(x_{0}, y_{0}) \in U \times V$, there exists $U_{0} \in \cn_{E}(x_{0})$, $V_{0} \in \cn_{F}(y_{0})$, and a unique $\alpha \in C^{k}(U_{0} \times V_{0}; V)$ such that

  1. (1)

    For every $y \in V_{0}$, $\alpha(x_{0}, y) = y$.

  2. (2)

    For every $(x, y) \in U_{0} \times V_{0}$,

    \[(\partial_{x} \alpha)(x, y) = f(x, \alpha(x, y))\]

Proof. Using translation, assume without loss of generality that $x_{0} = 0$ and $y_{0} = 0$. Let $B \in \cn_{E}(0)$ and

\[g: (-\eps, \eps) \times B \times V \to F \quad (t, z, y) \mapsto f(tz, y) \cdot z\]

then by Lemma 2.1.2, there exists $\delta > 0$, $B_{0} \in \cn_{E}(0)$, and $\beta: (-\delta, \delta) \times B_{0} \times V_{0}$ such that

  1. (a)

    For each $(z, y) \in B_{0} \times V_{0}$, $\beta(0, z, y) = y$.

  2. (b)

    For each $(t, z, y) \in (-\delta, \delta) \times B_{0} \times V_{0}$,

    \[\frac{d}{dt}\beta(t, z, y) = f(tz, \beta(t, z, y)) \cdot z\]

  3. (c)

    For each fixed $y \in V_{0}$, if $\beta(t, z) = \beta(t, z, y)$, then for each $(t, x) \in (-\delta, \delta) \times B_{0}$ and $h \in E$,

    \begin{align*}(\partial_{t}\partial_{z}\beta)(t, z)\cdot h&= t(\partial_{x}f)(tz, \beta(t, z))\cdot h \cdot z + f(tz, \beta(t, z))\cdot h \\&+ (\partial_{y}f)(tz, \beta(t, z)) \circ (\partial_{z} \beta)(t, z)\cdot h \cdot z\end{align*}

Following (c), let $k(t, z) = (\partial_{z}\beta)(t, z)\cdot h - tf(tz, \beta(t, z))\cdot h$, then

\begin{align*}\frac{d}{dt}k(t)&= (\partial_{t}\partial_{z}\beta)(t, z)\cdot h - f(tz, \beta)\cdot h \\&- t [(\partial_{x}f)(tz, \beta)]\cdot z \cdot h - t[(\partial_{y} f)(tz, \beta)]f(tz, \beta) \cdot z\cdot h\end{align*}

where by assumption,

\begin{align*}&[(\partial_{x}f)(tz, \beta)]\cdot z \cdot h - [(\partial_{y} f)(tz, \beta)]f(tz, \beta) \cdot z \cdot h \\&= Df(tz, \beta) \cdot (z, f(tz, \beta) \cdot z) \cdot h \\&= Df(tz, \beta) \cdot (h, f(tz, \beta) \cdot h) \cdot z \\&= [(\partial_{x}f)(tz, \beta)]\cdot h \cdot z - [(\partial_{y} f)(tz, \beta)]f(tz, \beta) \cdot h\cdot z\end{align*}

Therefore by (c),

\begin{align*}\frac{d}{dt}k(t)&= (\partial_{y}f)(tz, \beta) \circ (\partial_{z} \beta)(t, z)\cdot h \cdot z - t[(\partial_{y} f)(tz, \beta)]f(tz, \beta) \cdot h\cdot z \\&= (\partial_{y} f)(tz, \beta)[\partial_{z} \beta(t, z) - tf(tz, \beta)] \cdot h \cdot z \\&= (\partial_{y} f)(tz, \beta) \cdot k(t) \cdot z\end{align*}

Since $0$ is a solution to the above equation, $k(t) = 0$ by the uniqueness of solutions to ODEs. Hence for every $t \in (-\delta, \delta)$ and $z \in B_{0}$,

\[\partial_{z} \beta(t, z) = tf(tz, \beta(t, z))\]

By adjusting $\delta$ and $B_{0}$, assume without loss of generality that $\delta > 1$. In which case, if $\alpha(x) = \beta(1, x, y)$, then

\[\partial_{x}\alpha(x, y) = \partial_{z}\beta(1, x, y) = f(z, \beta(1, x, y)) = f(x, \alpha(x, y))\]

$\square$