基本定义
设\[h=\left(h_{1}, h_{2}, \cdots, h_{n}\right)\]
如果
\[f\left(x_{0}+h\right)-f\left(x_{0}\right)=\sum_{i=1}^{n} \lambda_{i} h_{i}+o(\|h\|), \quad\|h\| \rightarrow 0\]
其中\(\lambda_{1}, \lambda_{2}, \cdots, \lambda_{n}\)不依赖于h,那么f在\(x_0\)处可微,并记
\[\mathrm{d} f\left(\boldsymbol{x}_{0}\right)=\sum_{i=1}^{n} \lambda_{i} h_{i}\]
若f在D上每一点都可微,则称f是D上的可微函数,显然,微分是关于\(x_i\)的齐次线性函数
两式结合又可以写出\[f\left(x_{0}+h\right)-f\left(x_{0}\right)=d f\left(x_{0}\right)+o(\|h\|),\|h\| \rightarrow 0\]
\(\lambda_i\)的意义
接下来说明\({\lambda}_i\)的意义,在固定了\(n-1\)个变量后,对比偏导数或者方向导数,即可以发现,\(\lambda_i\)就是f在\(z_0\)处关于\(x_i\)的偏导数,也就是说
\[\lambda_{i}=D_{i} f\left(x_{0}\right)\]
同时也就有\[\operatorname{df}\left(x_{0}\right)=\sum_{i=1}^{n} \frac{\partial f\left(x_{0}\right)}{\partial x_{i}} h_{i}\]
可微推论
此时显然的结论是,可微等价推出f在\(z_0\)处一阶偏导数全部存在
且,可微能推出连续来。
这两个推论是不可逆的,由于例子
\(f(\boldsymbol{p})=\left\{\begin{array}{ll}{\frac{x y}{x^{2}+y^{2}},} & {p=(x, y) \neq(0,0)} \\ {0,} & {p=(0,0)}\end{array}\right.\)
Jacobi矩阵定义
定义函数(算子)
\[J f(x)=\left(D_{1} f(x), D_{2} f(x), \cdots, D_{n} f(x)\right)\]
并称它为f在x处的Jacobi矩阵,J也就相当于梯度算子
微分rewrite
至此,具备了引入矩阵来更加和谐的基础。
现在认为空间内的点均为列向量,
\(\boldsymbol{x}=\left(\begin{array}{c}{x_{1}} \\ {x_{2}} \\ {\vdots} \\ {x_{n}}\end{array}\right)\)
\(h=\left(\begin{array}{c}{h_{1}} \\ {h_{2}} \\ {\vdots} \\ {h_{h}}\end{array}\right)\)
可微的相关proposition
1.等价表述
在\(x_0\)处可微等价于对以下等式
\[f\left(x_{0}+h\right)-f\left(x_{0}\right)=J f\left(x_{0}\right) h+\sum_{i=1}^{n} \beta_{i}(h) h_{i}\]
当\(\|\boldsymbol{h}\| \rightarrow 0\)时,\[\beta_{i}(\boldsymbol{h}) \rightarrow 0, i=1,2, \cdots, n\]
2.充分条件
设开集\(D \subset \mathbf{R}^{n}\),函数\(f: \ D\to R\),If for all \(i\) from 1 to n and \(x_0 \in D\), \(D_i\) exists and is continuous in a neightboorhood of \(x_0\), then \(f\) is differentialle on \(x_0\)