【高等数学笔记】多元向量值函数的导数与微分

一、多元向量值函数导数与微分的定义

设有$n$元向量值函数$\mathbf{f}:U(x_0)\subseteq {\mathbb{R}}^n\to {\mathbb{R}}^m$，该函数可以表示成
$$\mathbf{f}(\mathbf{x})=\left[\begin{array}{c}{f}_1(\mathbf{x})\\ f_2(\mathbf{x})\\ \cdots \\ f_m(\mathbf{x})\end{array}\right]=\left[\begin{array}{c}{f}_1(x_1,x_2,x_3,\cdots ,x_n)\\ f_2(x_1,x_2,x_3,\cdots ,x_n)\\ \cdots \\ f_m(x_1,x_2,x_3,\cdots ,x_n)\end{array}\right]$$它将$n$维空间中的点$\mathbf{x}$映射为$m$维空间中的点$\mathbf{f}(\mathbf{x})$。若$\mathbf{f}$的每个分量$f_1,f_2,\cdots ,f_n$都在点${\mathbf{x}}_0$处可微，则我们定义$\mathbf{f}$在${\mathbf{x}}_0$处的导数（雅可比矩阵）为$$\mathrm{D}\mathbf{f}({\mathbf{x}}_0)=\left[\begin{array}{cccc}\frac{\partial f_1({\mathbf{x}}_0)}{\partial x_1}& \frac{\partial f_1({\mathbf{x}}_0)}{\partial x_2}& \cdots & \frac{\partial f_1({\mathbf{x}}_0)}{\partial x_n}\\ \frac{\partial f_2({\mathbf{x}}_0)}{\partial x_1}& \frac{\partial f_2({\mathbf{x}}_0)}{\partial x_2}& \cdots & \frac{\partial f_2({\mathbf{x}}_0)}{\partial x_n}\\ ⋮& ⋮& & ⋮\\ \frac{\partial f_m({\mathbf{x}}_0)}{\partial x_1}& \frac{\partial f_m({\mathbf{x}}_0)}{\partial x_2}& \cdots & \frac{\partial f_m({\mathbf{x}}_0)}{\partial x_n}\end{array}\right]=\left[\begin{array}{c}\nabla f_1({\mathbf{x}}_0)\\ \nabla f_2({\mathbf{x}}_0)\\ ⋮\\ \nabla f_m({\mathbf{x}}_0)\end{array}\right]$$定义$\mathbf{f}$在${\mathbf{x}}_0$处的微分为$$\begin{array}{rl}\text{d}\mathbf{f}({\mathbf{x}}_0)& =\left[\begin{array}{c}\text{d}{f}_1({\mathbf{x}}_0)\\ \text{d}{f}_2({\mathbf{x}}_0)\\ ⋮\\ \text{d}{f}_m({\mathbf{x}}_0)\end{array}\right]\\ & =\left[\begin{array}{c}\frac{\partial f_1({\mathbf{x}}_0)}{\partial x_1}\text{d}{x}_1+\frac{\partial f_1({\mathbf{x}}_0)}{\partial x_2}\text{d}{x}_2+\cdots +\frac{\partial f_1({\mathbf{x}}_0)}{\partial x_n}\text{d}{x}_n\\ \frac{\partial f_2({\mathbf{x}}_0)}{\partial x_1}\text{d}{x}_1+\frac{\partial f_2({\mathbf{x}}_0)}{\partial x_2}\text{d}{x}_2+\cdots +\frac{\partial f_2({\mathbf{x}}_0)}{\partial x_n}\text{d}{x}_n\\ ⋮\\ \frac{\partial f_m({\mathbf{x}}_0)}{\partial x_1}\text{d}{x}_1+\frac{\partial f_m({\mathbf{x}}_0)}{\partial x_2}\text{d}{x}_2+\cdots +\frac{\partial f_m({\mathbf{x}}_0)}{\partial x_n}\text{d}{x}_n\end{array}\right]\\ & =\left[\begin{array}{cccc}\frac{\partial f_1({\mathbf{x}}_0)}{\partial x_1}& \frac{\partial f_1({\mathbf{x}}_0)}{\partial x_2}& \cdots & \frac{\partial f_1({\mathbf{x}}_0)}{\partial x_n}\\ \frac{\partial f_2({\mathbf{x}}_0)}{\partial x_1}& \frac{\partial f_2({\mathbf{x}}_0)}{\partial x_2}& \cdots & \frac{\partial f_2({\mathbf{x}}_0)}{\partial x_n}\\ ⋮& ⋮& & ⋮\\ \frac{\partial f_m({\mathbf{x}}_0)}{\partial x_1}& \frac{\partial f_m({\mathbf{x}}_0)}{\partial x_2}& \cdots & \frac{\partial f_m({\mathbf{x}}_0)}{\partial x_n}\end{array}\right]\left[\begin{array}{c}\text{d}{x}_1\\ \text{d}{x}_2\\ \cdots \\ \text{d}{x}_n\end{array}\right]\\ & =\text{D}\mathbf{f}({\mathbf{x}}_0)\text{d}\mathbf{x}\end{array}$$当$m=n$时，雅可比矩阵为方阵，该方阵的行列式称为$\mathbf{f}$在${\mathbf{x}}_0$处的雅可比行列式（Jacobian Determinant，简称雅可比式），记作$${\mathbf{J}}_f({\mathbf{x}}_0)=\left|\mathrm{D}\mathbf{f}({\mathbf{x}}_0)\right|={\frac{\partial (f_1,f_2,\cdots ,f_n)}{\partial (x_1,x_2,\cdots ,x_n)}|}_{{\mathbf{x}}_0}$$例如，设$$\{\begin{array}{l}x(\rho ,\theta )=\rho \cos \theta \\ y(\rho ,\theta )=\rho \sin \theta \end{array}$$则$$\frac{\partial (x,y)}{\partial (\rho ,\theta )}=\left|\begin{array}{cc}\cos \theta & -\rho \sin \theta \\ \sin \theta & \rho \cos \theta \end{array}\right|=\rho ({\cos }^2\theta +{\sin }^2\theta )=\rho$$设${\mathbf{x}}_0=(x_1,x_2,\cdots ,x_n{)}^T$，我们定义$\mathbf{f}$在${\mathbf{x}}_0$处对$x_i$的偏导数为$$\frac{\partial \mathbf{f}({\mathbf{x}}_0)}{\partial x_i}=\underset{\Delta x_i\to 0}{\lim }\frac{\mathbf{f}\left(\left[\begin{array}{c}{x}_1\\ x_2\\ ⋮\\ x_i+\Delta x_i\\ ⋮\\ x_n\end{array}\right]\right)-\mathbf{f}\left(\left[\begin{array}{c}{x}_1\\ x_2\\ ⋮\\ x_i\\ ⋮\\ x_n\end{array}\right]\right)}{\Delta x_i}$$当$\frac{\partial \mathbf{f}({\mathbf{x}}_0)}{\partial x_i}$存在时，有$$\frac{\partial \mathbf{f}({\mathbf{x}}_0)}{\partial x_i}=\left[\begin{array}{c}\frac{\partial f_1({\mathbf{x}}_0)}{\partial x_i}\\ \frac{\partial f_2({\mathbf{x}}_0)}{\partial x_i}\\ ⋮\\ \frac{\partial f_m({\mathbf{x}}_0)}{\partial x_i}\end{array}\right]$$因此我们可以把雅可比矩阵$\mathrm{D}\mathbf{f}({\mathbf{x}}_0)$按列和按行分块分别写作$$\mathrm{D}\mathbf{f}({\mathbf{x}}_0)=\left[\begin{array}{cccc}\frac{\partial \mathbf{f}({\mathbf{x}}_0)}{\partial x_1}& \frac{\partial \mathbf{f}({\mathbf{x}}_0)}{\partial x_2}& \cdots & \frac{\partial \mathbf{f}({\mathbf{x}}_0)}{\partial x_n}\end{array}\right]=\left[\begin{array}{c}\nabla f_1({\mathbf{x}}_0)\\ \nabla f_2({\mathbf{x}}_0)\\ ⋮\\ \nabla f_m({\mathbf{x}}_0)\end{array}\right]$$

二、多元向量值函数的微分运算法则

定理1 设向量值函数$\mathbf{f}$和$\mathbf{g}$都在点$\mathbf{x}$处可微，$u$是在$\mathbf{x}$处可微的数量值函数，则
(1) $\mathbf{f}+\mathbf{g}$在$x$处可微，且其导数为$$\text{D}(\mathbf{f}+\mathbf{g})(\mathbf{x})=\mathrm{D}\mathbf{f}(\mathbf{x})+\mathrm{D}\mathbf{g}(\mathbf{x})$$(2) $⟨\mathbf{f},\mathbf{g}⟩$在$\mathbf{x}$处可微，且其导数为$$\mathrm{D}⟨\mathbf{f},\mathbf{g}⟩(\mathbf{x})=(\mathbf{f}(\mathbf{x}){)}^{\mathrm{T}}\mathrm{D}\mathbf{g}(\mathbf{x})+(\mathbf{g}(\mathbf{x}){)}^{\mathrm{T}}\mathrm{D}\mathbf{f}(\mathbf{x})$$(3) $u\mathbf{f}$在$\mathbf{x}$处可微，且其导数为$$\text{D}(u\mathbf{f})(\mathbf{x})=u(\mathbf{x})\text{D}\mathbf{f}(\mathbf{x})+\mathbf{f}(\mathbf{x})\nabla u(\mathbf{x})$$(4) 若$\mathbf{f}:\mathbb{R}\to {\mathbb{R}}^3,\mathbf{g}:\mathbb{R}\to {\mathbb{R}}^3$，则向量积$\mathbf{f}\times \mathbf{g}$在$x$处可微，且其导数为$$\text{D}(\mathbf{f}\times \mathbf{g})(x)=\text{D}\mathbf{f}(x)\times \mathbf{g}(x)+\mathbf{f}(x)\times \text{D}\mathbf{g}(x)$$证明：
(1) 显然成立。
(2) 设$\mathbf{f}=\left[\begin{array}{c}{f}_1\\ f_2\\ ⋮\\ f_m\end{array}\right],\mathbf{g}=\left[\begin{array}{c}{g}_1\\ g_2\\ ⋮\\ g_m\end{array}\right]$，则数量值函数$F=⟨\mathbf{f},\mathbf{g}⟩(\mathbf{x})=\sum _{i=1}^m{f}_i{g}_i$，且$$\begin{array}{rl}\text{D}F(\mathbf{x})& =\nabla F(\mathbf{x})\\ & =\sum _{i=1}^m\nabla f_i{g}_i(\mathbf{x})\end{array}$$我们知道，对于 j ∈ { 1 , 2 , ⋯   , n } j\in\{1,2,\cdots,n\} j∈{1,2,⋯,n}，$$\frac{\partial f_i{g}_i(x_j)}{\partial x_j}=f_i\frac{\partial g_i(x_j)}{\partial x_j}+g_i\frac{\partial f_i(x_j)}{\partial x_j}$$故$$\begin{array}{rl}\nabla f_i{g}_i(\mathbf{x})& =\left(f_i\frac{\partial g_i(x_1)}{\partial x_1}+g_i\frac{\partial f_i(x_1)}{\partial x_1},f_i\frac{\partial g_i(x_2)}{\partial x_2}+g_i\frac{\partial f_i(x_2)}{\partial x_2},\cdots ,f_i\frac{\partial g_i(x_n)}{\partial x_n}+g_i\frac{\partial f_i(x_n)}{\partial x_n}\right)\\ & =f_i(\mathbf{x})\nabla g_i(\mathbf{x})+\nabla f_i(\mathbf{x})g_i(\mathbf{x})\end{array}$$那么$$\begin{array}{rl}\text{D}F(\mathbf{x})& =\sum _{i=1}^m[f_i(\mathbf{x})\nabla g_i(\mathbf{x})+\nabla f_i(\mathbf{x})g_i(\mathbf{x})]\\ & =\left[\begin{array}{cccc}{f}_1(\mathbf{x})& f_2(\mathbf{x})& \cdots & f_m(\mathbf{x})\end{array}\right]\left[\begin{array}{c}\nabla g_1(\mathbf{x})\\ \nabla g_2(\mathbf{x})\\ ⋮\\ \nabla g_m(\mathbf{x})\end{array}\right]+\left[\begin{array}{cccc}{g}_1(\mathbf{x})& g_2(\mathbf{x})& \cdots & g_m(\mathbf{x})\end{array}\right]\left[\begin{array}{c}\nabla f_1(\mathbf{x})\\ \nabla f_2(\mathbf{x})\\ ⋮\\ \nabla f_m(\mathbf{x})\end{array}\right]\\ & =(\mathbf{f}(\mathbf{x}){)}^T\mathrm{D}\mathbf{g}(\mathbf{x})+(\mathbf{g}(\mathbf{x}){)}^{\mathrm{T}}\mathrm{D}\mathbf{f}(\mathbf{x})\end{array}$$
(3) 是(2)的特例。
(4) $$\begin{array}{rl}\text{D}(\mathbf{f}\times \mathbf{g})(x)& =\text{D}\left|\begin{array}{ccc}\mathbf{i}& \mathbf{j}& \mathbf{k}\\ f_1& f_2& f_3\\ g_1& g_2& g_3\end{array}\right|\\ & =\frac{\mathrm{d}}{\text{d}x}\left[\begin{array}{c}{f}_2{g}_3-f_3{g}_2\\ f_3{g}_1-f_1{g}_3\\ f_1{g}_2-f_2{g}_1\end{array}\right]\\ & =\left[\begin{array}{c}{f}_2^{\prime }{g}_3+f_2{g}_3^{\prime }-f_3^{\prime }{g}_2-f_3{g}_2^{\prime }\\ f_3^{\prime }{g}_1+f_3{g}_1^{\prime }-f_1^{\prime }{g}_3-f_1{g}_3^{\prime }\\ f_1^{\prime }{g}_2+f_1{g}_2^{\prime }-f_2^{\prime }{g}_1-f_2{g}_1^{\prime }\end{array}\right]\\ & =\left[\begin{array}{c}{f}_2^{\prime }{g}_3-f_3^{\prime }{g}_2\\ f_3^{\prime }{g}_1-f_1^{\prime }{g}_3\\ f_1^{\prime }{g}_2-f_2^{\prime }{g}_1\end{array}\right]+\left[\begin{array}{c}{f}_2{g}_3^{\prime }-f_3{g}_2^{\prime }\\ f_3{g}_1^{\prime }-f_1{g}_3^{\prime }\\ f_1{g}_2^{\prime }-f_2{g}_1^{\prime }\end{array}\right]\\ & =\left[\begin{array}{c}{f}_1^{\prime }\\ f_2^{\prime }\\ f_3^{\prime }\end{array}\right]\times \left[\begin{array}{c}{g}_1\\ g_2\\ g_3\end{array}\right]+\left[\begin{array}{c}{f}_1\\ f_2\\ f_3\end{array}\right]\times \left[\begin{array}{c}{g}_1^{\prime }\\ g_2^{\prime }\\ g_3^{\prime }\end{array}\right]\\ & =\text{D}\mathbf{f}(x)\times \mathbf{g}(x)+\mathbf{f}(x)\times \text{D}\mathbf{g}(x)\end{array}$$证毕。∎

定理2 设$\mathbf{r}=\mathbf{r}(t)$表示空间中动点$\left[\begin{array}{c}x(t)\\ y(t)\\ z(t)\end{array}\right]$的向径，则$$⟨{\mathbf{r}}^{\prime }(t),\mathbf{r}(t)⟩\equiv 0⟺\Vert \mathbf{r}(t)\Vert \equiv c$$其中$c$为常数，这表示动点的轨迹在以原点为中心的球面上。
证明：
由定理1(2)知$$\frac{\mathrm{d}}{\text{d}t}⟨\mathbf{r},\mathbf{r}⟩(t)=⟨\mathbf{r},{\mathbf{r}}^{\prime }⟩(t)+⟨{\mathbf{r}}^{\prime },\mathbf{r}⟩(t)$$即$$\frac{\mathrm{d}}{\text{d}t}\Vert \mathbf{r}(t){\Vert }^2=2⟨{\mathbf{r}}^{\prime }(t),\mathbf{r}(t)⟩$$故$⟨{\mathbf{r}}^{\prime }(t),\mathbf{r}(t)⟩\equiv 0⟺\frac{\mathrm{d}}{\text{d}t}\Vert \mathbf{r}(t){\Vert }^2\equiv 0⟺\Vert \mathbf{r}(t){\Vert }^2\equiv C⟺\Vert \mathbf{r}(t)\Vert \equiv c$。∎

定理3（向量值函数的链式法则） 设向量值函数$\mathbf{g}(\mathbf{x}):{\mathbb{R}}^n\to {\mathbb{R}}^p$在点${\mathbf{x}}_0$处可微，向量值函数$\mathbf{f}(\mathbf{u}):{\mathbb{R}}^p\to {\mathbb{R}}^m$在点$\mathbf{g}({\mathbf{x}}_0)$处可微，则复合函数$\mathbf{w}(\mathbf{x})=\mathbf{f}(\mathbf{g}(\mathbf{x}))$在点${\mathbf{x}}_0$处可微，且$$\mathrm{D}\mathbf{w}({\mathbf{x}}_0)=\mathrm{D}\mathbf{f}(\mathbf{g}({\mathbf{x}}_0))\mathrm{D}\mathbf{g}({\mathbf{x}}_0)$$证明提要：由复合函数的求导法则，$\mathrm{D}\mathbf{w}(\mathbf{x})$的第$(i,j)$个元素$\frac{\partial w_i}{\partial x_j}$可以表示为$$\begin{array}{rl}\frac{\partial w_i}{\partial x_j}& =\frac{\partial f_i(\mathbf{g}(\mathbf{x}))}{\partial x_j}\\ & =\frac{\partial f_i(g_1(x_1,x_2,\cdots ,x_n),g_2(x_1,x_2,\cdots ,x_n),\cdots ,g_p(x_1,x_2,\cdots ,x_n))}{\partial x_j}\\ & =\sum _{k=1}^p\frac{\partial f_i}{\partial u_k}\frac{\partial g_k}{\partial x_j}\end{array}$$恰好是矩阵乘法的形式。∎

当$n=p=m$时，取两端的行列式得$$\frac{\partial (w_1,w_2,\cdots ,w_n)}{\partial (x_1,x_2,\cdots ,x_n)}=\frac{\partial (w_1,w_2,\cdots ,w_n)}{\partial (u_1,u_2,\cdots ,u_n)}\frac{\partial (g_1,g_2,\cdots ,g_n)}{\partial (x_1,x_2,\cdots ,x_n)}$$即复合函数的雅可比式是两个函数的雅可比式的乘积（${\mathbf{J}}_w={\mathbf{J}}_f{\mathbf{J}}_g$）。

推论 设$\mathbf{f},\mathbf{g}:{\mathbb{R}}^n\to {\mathbb{R}}^n$互为反函数，则它们的雅可比式互为倒数。
证明：令$\mathbf{w}(\mathbf{x})=\mathbf{f}(\mathbf{g}(\mathbf{x}))\equiv \mathbf{x}$，而$\mathrm{D}\mathbf{w}(\mathbf{x})=\mathbf{I}$，故${\mathbf{J}}_w=1$，${\mathbf{J}}_f{\mathbf{J}}_g=1$。∎

三、由方程所确定的隐函数的微分法

例1 已知方程组$\{\begin{array}{l}{F}_1(x,y,u,v)=xu+yv=0\\ F_2(x,y,u,v)=yu+xv=1\end{array}$确定了隐函数$\{\begin{array}{l}u=u(x,y)\\ v=v(x,y)\end{array}$，求(1)$\frac{\partial u}{\partial x}$及(2)$\frac{\partial v}{\partial y}$。
解：
(1) 方程组两边对$x$求偏导得$$\{\begin{array}{l}u+x\frac{\partial u}{\partial x}+y\frac{\partial v}{\partial x}=0\\ y\frac{\partial u}{\partial x}+v+x\frac{\partial v}{\partial x}=0\end{array}$$这是一个关于$\frac{\partial u}{\partial x},\frac{\partial v}{\partial x}$的二元一次方程组，可以表示成$$\left[\begin{array}{cc}x& y\\ y& x\end{array}\right]\left[\begin{array}{c}\frac{\partial u}{\partial x}\\ \frac{\partial v}{\partial x}\end{array}\right]=\left[\begin{array}{c}-u\\ -v\end{array}\right]$$其中系数行列式即为雅可比式$\mathbf{J}=\frac{\partial (F_1,F_2)}{\partial (u,v)}$。根据Cramer法则，$$\frac{\partial u}{\partial x}=\frac{{\mathbf{J}}_u}{\mathbf{J}}=\frac{\left|\begin{array}{cc}-u& y\\ -v& x\end{array}\right|}{\left|\begin{array}{cc}x& y\\ y& x\end{array}\right|}=\frac{vy-ux}{x^2-y^2}$$(2) 同理可得$\frac{\partial v}{\partial y}=\frac{vy-ux}{x^2-y^2}$。注意，原方程中将$u$与$v$互换、$x$与$y$互换所得的方程与原方程完全相同，所以$\frac{\partial u}{\partial x}=\frac{\partial v}{\partial y}$。∎

例2 已知$\{\begin{array}{l}u+v+w=x\\ uv+vw+wu=y\\ uvw=z\end{array}$，求$\frac{\partial u}{\partial x},\frac{\partial u}{\partial y},\frac{\partial u}{\partial z}$。
解：两边取微分得$$\{\begin{array}{ll}\text{d}u+\text{d}v+\text{d}w& =\text{d}x\\ (v+w)\text{d}u+(u+w)\text{d}v+(u+v)\text{d}w& =\text{d}y\\ vw\text{d}u+uw\text{d}v+uv\text{d}w& =\text{d}z\end{array}$$即$$\left[\begin{array}{ccc}1& 1& 1\\ v+w& u+w& u+v\\ vw& uw& uv\end{array}\right]\left[\begin{array}{c}\text{d}u\\ \text{d}v\\ \text{d}w\end{array}\right]=\left[\begin{array}{c}\text{d}x\\ \text{d}y\\ \text{d}z\end{array}\right]$$雅可比行列式$$\begin{array}{rl}\mathbf{J}& =\left|\begin{array}{ccc}1& 1& 1\\ v+w& u+w& u+v\\ vw& uw& uv\end{array}\right|\\ & =u^2(v-w)+v^2(w-u)+w^2(u-v)\\ & =(u-v)(v-w)(u-w)\end{array}$$故$$\begin{array}{rl}\text{d}u& =\frac{{\mathbf{J}}_u}{\mathbf{J}}\\ & =\frac{\left|\begin{array}{ccc}\text{d}x& 1& 1\\ \text{d}y& u+w& u+v\\ \text{d}z& uw& uv\end{array}\right|}{\mathbf{J}}\\ & =\frac{u^2(v-w)\text{d}x-u(v-w)\text{d}y+(v-w)\text{d}z}{\mathbf{J}}\\ & =\frac{u^2\text{d}x-u\text{d}y+\text{d}z}{(u-v)(u-w)}\end{array}$$因此$$\{\begin{array}{l}\frac{\partial u}{\partial x}=\frac{u^2}{(u-v)(u-w)}\\ \frac{\partial u}{\partial y}=-\frac{u}{(u-v)(u-w)}\\ \frac{\partial u}{\partial z}=\frac{1}{(u-v)(u-w)}\end{array}$$∎