必要性:因为\(\varphi\)是满射,所以\(V=\varphi(V)\)。又\(V=V_1+V_2\),故
\begin{equation*}
V=\varphi(V_1+V_2)=\varphi(V_1)+\varphi(V_2).
\end{equation*}
对任意\(\beta\in\varphi(V_1)\bigcap \varphi(V_2)\),存在\(\alpha_1\in V_1,\alpha_2\in V_2\)使得\(\beta=\varphi(\alpha_1),\beta=\varphi(\alpha_2)\)。由\(\varphi\)是单射知\(\alpha_1=\alpha_2\in V_1\bigcap V_2\),注意到\(V_1+V_2\)是直和,所以\(\alpha_1=\alpha_2=0\),于是\(\beta=\varphi(\alpha_1)=0\),因此\(\varphi(V_1)\bigcap \varphi(V_2)=\{0\}\)。综上,\(V =\varphi(V_1)\oplus \varphi(V_2)\)。
充分性:由\(V=V_1+V_2\)得\(\varphi(V)=\varphi(V_1)+\varphi(V_2)\),又\(V=\varphi(V_1)+\varphi(V_2)\),故\(V=\varphi(V)\),即\(\varphi\)是满射。下证\(\varphi\)是单射,即证由\(\varphi(\alpha)=0\)可推出\(\alpha=0\)。
证法一:设\(\xi_1,\cdots ,\xi_r\)是\(V_1\)的一个基,\(\xi_{r+1},\cdots ,\xi_n\)是\(V_2\)的一个基,则\(\xi_1,\cdots ,\xi_n\)是\(V\)的一个基且
\begin{equation*}
\varphi(V_1)=\langle\varphi(\xi_1),\cdots,\varphi(\xi_r)\rangle,\varphi(V_2)=\langle\varphi(\xi_{r+1}),\cdots,\varphi(\xi_n)\rangle.
\end{equation*}
于是\(\dim\varphi(V_1)\leq r,\dim\varphi(V_2)\leq n-r\)。由\(V=\varphi(V_1)\oplus \varphi(V_2)\)知\(\dim\varphi(V_1)+\dim\varphi(V_2)=n\),所以\(\dim\varphi(V_1)= r\)且\(\dim\varphi(V_2)= n-r\)。于是\(\varphi(\xi_1),\cdots ,\varphi(\xi_r)\)是\(\varphi(V_1)\)的一个基, \(\varphi(\xi_{r+1}),\cdots,\varphi(\xi_n)\)是\(\varphi(V_2)\)的一个基,\(\varphi(\xi_1),\cdots ,\varphi(\xi_n)\)是\(V\)的一个基。
设\(\varphi(\alpha)=0\),由于\(\xi_1,\cdots ,\xi_n\)是\(V\)的一个基,所以存在\(c_1,\cdots ,c_n\in\mathbb{F}\),使得\(\alpha=\sum\limits_{i=1}^n c_i \xi_i\)。于是
\begin{equation*}
0=\varphi(\alpha)=\sum\limits_{i=1}^n c_i \varphi(\xi_i),
\end{equation*}
由\(\varphi(\xi_1),\cdots ,\varphi(\xi_n)\)线性无关得
\begin{equation*}
c_1=\cdots =c_n=0,
\end{equation*}
故\(\alpha=0\),即\(\varphi\)是单射。因此\(\varphi\)是可逆变换。
证法二:(反证法)假设存在非零向量\(\alpha\in V\),使得\(\varphi(\alpha)=0\)。由\(V=V_1\oplus V_2\)知:存在\(\alpha_1\in V_1,\alpha_2\in V_2\),使得\(\alpha=\alpha_1+\alpha_2\),这里\(\alpha_1,\alpha_2\)不全为零向量。不妨设\(\alpha_1\neq 0\),则可将其扩充为\(V_1\)的一个基\(\alpha_1,\beta_2,\cdots ,\beta_r\)。由4、(3)结论得
\begin{equation*}
\varphi(V_1)=\varphi(\langle\alpha_1,\beta_2,\cdots ,\beta_r\rangle)=\langle\varphi(\alpha_1),\varphi(\beta_2)\cdots,\varphi(\beta_r)\rangle,
\end{equation*}
则\(\dim \varphi(V_1)\leq r=\dim V_1\)。同理,\(\dim\varphi(V_2)\leq\dim V_2\)。由于
\begin{equation*}
0=\varphi(\alpha)=\varphi(\alpha_1)+\varphi(\alpha_2),
\end{equation*}
这里\(\varphi(\alpha_1)\in\varphi(V_1),\varphi(\alpha_2)\in\varphi(V_2)\),又\(\varphi(V_1)+\varphi(V_2)\)是直和,所以
\begin{equation*}
\varphi(\alpha_1)=\varphi(\alpha_2)=0.
\end{equation*}
则\(\varphi(V_1)=\langle\varphi(\beta_2)\cdots,\varphi(\beta_r)\rangle\),即\(\dim\varphi (V_1)<r=\dim V_1\)。于是
\begin{equation*}
\dim\varphi(V_1)+\dim\varphi(V_2)<\dim V_1+\dim V_2,
\end{equation*}
这与\(V=V_1\oplus V_2\)且\(V=\varphi(V_1)\oplus \varphi(V_2)\)相矛盾。因此\(\varphi\)是单射。从而\(\varphi\)是可逆变换。