偏导数和全微分

偏导数

P0(x0,y0)R2P_0(x_0, y_0) \in \R^2,函数 z=f(x,y)z = f(x, y)P0P_0δ\delta 邻域 Nδ(P0)N_{\delta}(P_0) 内有定义,在 Nδ(P0)N_{\delta}(P_0) 中固定 y=y0y = y_0,得到一个关于 xx 的函数 f(x,y0)f(x, y_0),若 f(x,y0)f(x, y_0)x=x0x = x_0 处可导,即

limΔx0f(x0+Δx,y0)f(x0,y0)Δx\lim_{\Delta x \to 0} \frac{f(x_0 + \Delta x, y_0) - f(x_0, y_0)}{\Delta x}

存在,则称函数 z=f(x,y)z = f(x, y) 在点 (x0,y0)(x_0, y_0) 处对 xx偏导数存在,记作

fx(x0,y0),f1(x0,y0),fx(x0,y0),fx(x0,y0),zx(x0,y0),zx(x0,y0)f'_x(x_0, y_0),\\ f'_1(x_0, y_0),\\ \frac{\partial f}{\partial x}(x_0, y_0),\\ \frac{\partial f}{\partial x}\as_{\left(x_0, y_0\right)},\\ \frac{\partial z}{\partial x}\left(x_0, y_0\right),\\ \frac{\partial z}{\partial x}\as_{\left(x_0, y_0\right)}

同理可定义函数 z=f(x,y)z = f(x, y) 在点 (x0,y0)(x_0, y_0) 处对 yy偏导数

若二元函数 f(x,y)f(x, y) 在点 (x0,y0)(x_0, y_0) 处对 x,yx,\, y 的偏导数均存在,则称 f(x,y)f(x, y) 在点 (x0,y0)(x_0, y_0)可偏导。若 f(x,y)f(x, y)GG 中每一点都可偏导,则称 f(x,y)f(x, y)GG可偏导

(x0,y0)G(x_0, y_0) \in G,函数 z=f(x,y)z = f(x, y)(x0,y0)(x_0, y_0) 处对 xx 的偏导数记为

fx(x0,y0),fx,f1,fx,zxf'_x(x_0, y_0),\\ f'_x,\\ f'_1,\\ \frac{\partial f}{\partial x},\\ \frac{\partial z}{\partial x}

(x0,y0)(x_0, y_0) 处对 yy 的偏导数、nn 元函数的偏导数等类似。

偏导数的几何意义,可以看作是多元函数在某一方向上的变化率。

例如,对于二元函数 z=f(x,y)z = f(x, y) 在空间中为一个曲面,z=f(x,y0)z = f(x, y_0) 表示 zz 与平面 y=y0y = y_0 的交线,也就是说,ffxx 的偏导数 fx(x0,y0)f'_x(x_0, y_0) 表示 P(x0,y0,f(x0,y0))P(x_0, y_0, f(x_0, y_0)) 点处切线对 xx 轴的斜率,这条切线显然与 xOzx O z 平面平行。

设函数 z=f(x,y)z = f(x, y)P0(x0,y0)P_0 (x_0, y_0) 的某邻域 Nδ(P0)N_{\delta}(P_0) 内可偏导,且 fx(x,y),fy(x,y)f'_x(x, y),\, f'_y(x, y)Nδ(P0)N_{\delta}(P_0) 内有界,则函数 z=f(x,y)z = f(x, y)P0P_0 处连续。


证明:

Δz=f(x0+Δx,y0+Δy)f(x0,y0)=(f(x,y)f(x0,y))+(f(x0,y)f(x0,y0))\begin{aligned} \Delta z &= f(x_0 + \Delta x, y_0 + \Delta y) - f(x_0, y_0)\\ &= \Bigl(f(x, y) - f(x_0, y)\Bigr) + \Bigl(f(x_0, y) - f(x_0, y_0)\Bigr)\\ \end{aligned}

由一元函数微分中值定理知存在 ξ,η\xi,\, \eta 分别介于 xxx0x_0yyy0y_0 之间,使得

Δz=fx(ξ,y)(xx0)+fy(x0,η)(yy0)\Delta z = f'_x(\xi, y) (x - x_0) + f'_y(x_0, \eta) (y - y_0)

由于 fx(x,y),fy(x,y)f'_x(x, y),\, f'_y(x, y)Nδ(P0)N_{\delta}(P_0) 内有界,故

limxx0yy0Δz=0\lim_{\substack{x \to x_0\\ y \to y_0}} \Delta z = 0

f(x,y)f(x, y)P0P_0 处连续。

不同于一元函数,对于多元函数,即使其在某点处可偏导,也不一定在该点处连续。

例如函数

f(x,y)={xyx2+y2,(x,y)(0,0)0,(x,y)=(0,0)f(x, y) = \begin{cases} \dfrac{xy}{x^2 + y^2}, & (x, y) \ne (0, 0)\\ 0, & (x, y) = (0, 0) \end{cases}

fx(0,0)=limx0f(x,0)f(0,0)x0=limx000x=0fy(0,0)=limy0f(0,y)f(0,0)y0=limy000y=0\begin{aligned} f'_x(0, 0) &= \lim\limits_{x \to 0} \frac{f(x, 0) - f(0, 0)}{x - 0} &= \lim\limits_{x \to 0} \frac{0 - 0}{x} &= 0\\ f'_y(0, 0) &= \lim\limits_{y \to 0} \frac{f(0, y) - f(0, 0)}{y - 0} &= \lim\limits_{y \to 0} \frac{0 - 0}{y} &= 0 \end{aligned}

f(x,y)f(x, y)(0,0)(0, 0) 处可偏导,但

limx0y0f(x,y)\lim_{\substack{x \to 0\\ y \to 0}} f(x, y)

不存在,故 f(x,y)f(x, y)(0,0)(0, 0) 处不连续。

极限不存在的证明:

(x,y)(x, y) 沿着 y=xy = x 趋于 (0,0)(0, 0) 时,有

limx0y0f(x,y)=limx0x2x2+x2=12\begin{aligned} \lim_{\substack{x \to 0\\ y \to 0}} f(x, y) &= \lim_{x \to 0} \frac{x^2}{x^2 + x^2}\\ &= \frac{1}{2} \end{aligned}

(x,y)(x, y) 沿着 y=0y = 0 趋于 (0,0)(0, 0) 时,有

limx0y0f(x,y)=limx00x2=0\begin{aligned} \lim_{\substack{x \to 0\\ y \to 0}} f(x, y) &= \lim_{x \to 0} \frac{0}{x^2}\\ &= 0 \end{aligned}

两个极限不相等,故 f(x,y)f(x, y)(0,0)(0, 0) 处不存在极限。

而连续也无法推出可偏导,这点比较显然,毕竟一元函数连续无法推出可导。

类似于一元函数的高阶导数,多元函数也有高阶偏导数的概念。由于多元函数的自变量有多个,因此高阶偏导数有多种求法,例如对于二元函数 z=f(x,y)z = f(x, y),可以先对 xx 求偏导,再对 yy 求偏导,也可以先对 yy 求偏导,再对 xx 求偏导,还可以对 xx 求两次偏导,也可以对 yy 求两次偏导,等等。具体求法与一元函数类似,这里不再赘述。

若二阶混合偏导数 fxy(x,y)f''_{xy}(x, y)fyx(x,y)f''_{yx}(x, y) 在点 (x,y)(x, y)连续,则

fxy(x,y)=fyx(x,y)f''_{xy}(x, y) = f''_{yx}(x, y)

即混合偏导数与求导次序无关,有

2fxy=2fyx\frac{\partial^2 f}{\partial x \partial y} = \frac{\partial^2 f}{\partial y \partial x}


证明:

记辅助函数

F(h,k)=f(x+h,y+k)f(x+h,y)f(x,y+k)+f(x,y)F(h, k) = f(x + h, y + k) - f(x + h, y) - f(x, y + k) + f(x, y)

其中 h,k\left\lvert h \right\rvert,\, \left\lvert k \right\rvert 充分小时,令

φ(X)=f(X,y+k)f(X,y)\varphi (X) = f(X, y + k) - f(X, y)

F(h,k)=φ(x+h)φ(x)F(h, k) = \varphi (x + h) - \varphi (x)

运用一元函数的拉格朗日中值定理,存在 θ1,θ2(0,1)\theta_1,\, \theta_2 \in (0, 1),使得

F(h,k)=φ(x+θ1h)h=(fx(x+θ1h,y+k)fx(x+θ1h,y))h=fxy(x+θ1h,y+θ2k)hk\begin{aligned} F(h, k) &= \varphi'(x + \theta_1 h) h\\ &= \Bigl( f'_x(x + \theta_1 h, y + k) - f'_x(x + \theta_1 h, y) \Bigr) h\\ &= f''_{xy}(x + \theta_1 h, y + \theta_2 k) h k \end{aligned}

同理令

ψ(Y)=f(x+h,Y)f(x,Y)\psi (Y) = f(x + h, Y) - f(x, Y)

又存在 θ3,θ4(0,1)\theta_3,\, \theta_4 \in (0, 1),使得

F(h,k)=ψ(y+θ3k)k=(fy(x+h,y+θ3k)fy(x,y+θ3k))k=fyx(x+θ4h,y+θ3k)hk\begin{aligned} F(h, k) &= \psi'(y + \theta_3 k) k\\ &= \Bigl( f'_y(x + h, y + \theta_3 k) - f'_y(x, y + \theta_3 k) \Bigr) k\\ &= f''_{yx}(x + \theta_4 h, y + \theta_3 k) h k \end{aligned}

fxy(x+θ1h,y+θ2k)=fyx(x+θ4h,y+θ3k)f''_{xy}(x + \theta_1 h, y + \theta_2 k) = f''_{yx}(x + \theta_4 h, y + \theta_3 k)

h0,k0h \to 0,\, k \to 0,由于 fxy(x,y),fyx(x,y)f''_{xy}(x, y),\, f''_{yx}(x, y) 在点 (x,y)(x, y) 处连续,故

fxy(x,y)=fyx(x,y)f''_{xy}(x, y) = f''_{yx}(x, y)

上面的结论可以推广到 nn 元函数的偏导数上,只需记为分式形式即可。

全微分

设函数 z=f(x,y)z = f(x, y) 在点 P(x,y)P(x, y) 的某邻域内有定义,若函数 z=f(x,y)z = f(x, y) 在点 PP 的全增量

Δz=f(x+Δx,y+Δy)f(x,y)\Delta z = f(x + \Delta x, y + \Delta y) - f(x, y)

可表示为

Δz=AΔx+BΔy+o(ρ)\Delta z = A \Delta x + B \Delta y + o(\rho)

其中 A,BA,\, B 只与 PP 有关,而与 Δx,Δy\Delta x,\, \Delta y 无关,且 ρ=(Δx)2+(Δy)2\rho = \sqrt{(\Delta x)^2 + (\Delta y)^2}o(ρ)o(\rho) 是当 ρ0+\rho \to 0^{+} 时比 ρ\rho 高阶的无穷小,则称函数 f(x,y)f(x, y) 在点 (x,y)(x, y) 处可微,其线性部分 AΔx+BΔyA \Delta x + B \Delta y 称为函数 z=f(x,y)z = f(x, y) 在点 (x,y)(x, y) 处的全微分,记作

 ⁣dz=AΔx+BΔy\d z = A \Delta x + B \Delta y

设函数 z=f(x,y)z = f(x, y)(x,y)(x, y) 处可微,则函数 f(x,y)f(x, y)(x,y)(x, y) 处连续。


证明:

limΔx0Δy0Δz=limΔx0Δy0(AΔx+BΔy+o(ρ))=0\begin{aligned} \lim_{\substack{\Delta x \to 0\\\Delta y \to 0}} \Delta z &= \lim_{\substack{\Delta x \to 0\\\Delta y \to 0}} \left(A \Delta x + B \Delta y + o(\rho)\right)\\ &= 0 \end{aligned}

f(x,y)f(x, y)(x,y)(x, y) 处连续。

设函数 z=f(x,y)z = f(x, y)(x,y)(x, y) 处可微,则

 ⁣dz= ⁣z ⁣x ⁣dx+ ⁣z ⁣y ⁣dy\d z = \dfrac{\pd z}{\pd x} \d x + \dfrac{\pd z}{\pd y} \d y


证明:

Δy=0\Delta y = 0,得

ΔzΔx=A+o(ρ)Δx\dfrac{\Delta z}{\Delta x} = A + \dfrac{o(\rho)}{\Delta x}

从而

 ⁣z ⁣x=limΔx0ΔzΔx=A+limΔx0o(Δx)Δx=A\begin{aligned} \dfrac{\pd z}{\pd x} &= \lim_{\Delta x \to 0} \dfrac{\Delta z}{\Delta x}\\ &= A + \lim_{\Delta x \to 0} \dfrac{o(\left\lvert \Delta x \right\rvert)}{\Delta x}\\ &= A \end{aligned}

同理可得  ⁣z ⁣y=B\dfrac{\pd z}{\pd y} = B,故

 ⁣dz=A ⁣dx+B ⁣dy= ⁣z ⁣xΔx+ ⁣z ⁣yΔy\begin{aligned} \d z &= A \d x + B \d y\\ &= \dfrac{\pd z}{\pd x} \Delta x + \dfrac{\pd z}{\pd y} \Delta y \end{aligned}

类似一元函数微分的证明,分别取函数 f(x,y)=xf(x, y) = xf(x,y)=yf(x, y) = y,可得

 ⁣dx=Δx, ⁣dy=Δy\d x = \Delta x,\quad \d y = \Delta y

从而

 ⁣dz= ⁣z ⁣x ⁣dx+ ⁣z ⁣y ⁣dy\d z = \dfrac{\pd z}{\pd x} \d x + \dfrac{\pd z}{\pd y} \d y

由上面的内容可知,连续和可偏导是可微的必要条件,但不是充分条件。

f(x,y)={xyx2+y2,(x,y)(0,0)0,(x,y)=(0,0)f(x, y) = \begin{cases} \dfrac{xy}{\sqrt{x^2 + y^2}}, & (x, y) \ne (0, 0)\\ 0, & (x, y) = (0, 0) \end{cases},可以证明 f(x,y)f(x, y)(0,0)(0, 0) 处连续、可偏导,但不可微。

x=ρcosθ,y=ρsinθx = \rho \cos \theta,\, y = \rho \sin \theta,从而

limx0y0f(x,y)=limρ0+ρ2cosθsinθρ=0=f(0,0)\begin{aligned} \lim_{\substack{x \to 0\\ y \to 0}} f(x, y) &= \lim_{\rho \to 0^{+}} \frac{\rho^2 \cos \theta \sin \theta}{\rho}\\ &= 0\\ &= f(0, 0) \end{aligned}

从而 f(x,y)f(x, y)(0,0)(0, 0) 处连续。

 ⁣z ⁣x(0,0)=limx0f(x,0)f(0,0)x=limx00x=0\begin{aligned} \dfrac{\pd z}{\pd x}(0, 0) &= \lim\limits_{x \to 0} \dfrac{f(x, 0) - f(0, 0)}{x}\\ &= \lim\limits_{x \to 0} \dfrac{0}{x}\\ &= 0 \end{aligned}

 ⁣z ⁣y(0,0)=limy0f(0,y)f(0,0)y=limy00y=0\begin{aligned} \dfrac{\pd z}{\pd y}(0, 0) &= \lim\limits_{y \to 0} \dfrac{f(0, y) - f(0, 0)}{y}\\ &= \lim\limits_{y \to 0} \dfrac{0}{y}\\ &= 0 \end{aligned}

从而 f(x,y)f(x, y)(0,0)(0, 0) 处可偏导。

则全增量(由 Δz= ⁣z ⁣xΔx+ ⁣z ⁣yΔy+o(ρ)\Delta z = \dfrac{\pd z}{\pd x}\Delta x + \dfrac{\pd z}{\pd y}\Delta y + o(\rho),其中 ρ=(Δx)2+(Δy)2\rho = \sqrt{(\Delta x)^2 + (\Delta y)^2}。此处 Δx=x0=x,Δy=y0=y\Delta x = x - 0 = x,\, \Delta y = y - 0 = y)有

Δz=f(x,y)f(0,0)=x ⁣z ⁣x(0,0)+y ⁣z ⁣y(0,0)+ω=ω\begin{aligned} \Delta z &= f(x, y) - f(0, 0)\\ &= x\dfrac{\pd z}{\pd x}(0, 0) + y\dfrac{\pd z}{\pd y}(0, 0) + \omega\\ &= \omega \end{aligned}

从而

ω=f(x,y)=xyx2+y2\omega = f(x, y) = \dfrac{xy}{\sqrt{x^2 + y^2}}

f(x,y)f(x, y)(0,0)(0, 0) 处可微,则 ω\omega 应该为 ρ=x2+y2\rho = \sqrt{x^2 + y^2} 的高阶无穷小,但

ωρ=ρcosθsinθρ2=cosθsinθ0\begin{aligned} \dfrac{\omega}{\rho} &= \dfrac{\rho \cos \theta \sin \theta}{ \rho^2}\\ &= \cos \theta \sin \theta \nrightarrow 0 \end{aligned}

从而 f(x,y)f(x, y)(0,0)(0, 0) 处不可微。

设函数 z=f(x,y)z = f(x, y) 在点 (x,y)(x, y) 的某邻域内可偏导,且  ⁣z ⁣x\dfrac{\pd z}{\pd x} ⁣z ⁣y\dfrac{\pd z}{\pd y} 在点 (x,y)(x, y) 连续,则 f(x,y)f(x, y) 在点 (x,y)(x, y) 可微。(函数可微的充分条件


证明:

考虑 z=f(x,y)z = f(x, y) 在点 (x,y)(x, y) 处的全增量

Δz=f(x+Δx,y+Δy)f(x,y)=(f(x+Δx,y+Δy)f(x,y+Δy))+(f(x,y+Δy)f(x,y))\begin{aligned} \Delta z &= f(x + \Delta x, y + \Delta y) - f(x, y)\\ &= \Bigr(f(x + \Delta x, y + \Delta y) - f(x, y + \Delta y)\Bigr) + \Bigr(f(x, y + \Delta y) - f(x, y)\Bigr)\\ \end{aligned}

Δx,Δy\left\lvert \Delta x \right\rvert,\, \left\lvert \Delta y \right\rvert 充分小时,由一元函数拉格朗日中值定理,存在 θ1,θ2(0,1)\theta_1,\, \theta_2 \in (0, 1),使得

Δz=Δx ⁣z ⁣x(x+θ1Δx,y+Δy)+Δy ⁣z ⁣y(x,y+θ2Δy)\begin{aligned} \Delta z = & \Delta x\dfrac{\pd z}{\pd x}(x + \theta_1 \Delta x, y +\Delta y) + \Delta y\dfrac{\pd z}{\pd y}(x, y + \theta_2 \Delta y)\\ \end{aligned}

由于  ⁣z ⁣x\dfrac{\pd z}{\pd x} ⁣z ⁣y\dfrac{\pd z}{\pd y} 在点 (x,y)(x, y) 连续,从而

limΔx0Δy0 ⁣z ⁣x(x+θ1Δx,y+Δy)= ⁣z ⁣x(x,y)limΔx0Δy0 ⁣z ⁣y(x,y+θ2Δy)= ⁣z ⁣y(x,y)\begin{aligned} \lim_{\substack{\Delta x \to 0\\ \Delta y \to 0}} \dfrac{\pd z}{\pd x}(x + \theta_1 \Delta x, y +\Delta y) &= \dfrac{\pd z}{\pd x}(x, y)\\ \lim_{\substack{\Delta x \to 0\\ \Delta y \to 0}} \dfrac{\pd z}{\pd y}(x, y + \theta_2 \Delta y) &= \dfrac{\pd z}{\pd y}(x, y)\\ \end{aligned}

所以

 ⁣z ⁣x(x+θ1Δx,y+Δy)= ⁣z ⁣x(x,y)+α ⁣z ⁣y(x,y+θ2Δy)= ⁣z ⁣y(x,y)+β\begin{aligned} \dfrac{\pd z}{\pd x}(x + \theta_1 \Delta x, y +\Delta y) &= \dfrac{\pd z}{\pd x}(x, y) + \alpha\\ \dfrac{\pd z}{\pd y}(x, y + \theta_2 \Delta y) &= \dfrac{\pd z}{\pd y}(x, y) + \beta\\ \end{aligned}

其中 α,β0(Δx0,Δy0)\alpha,\, \beta \to 0\, (\Delta x \to 0,\, \Delta y \to 0)

从而

Δz=Δx ⁣z ⁣x(x,y)+Δy ⁣z ⁣y(x,y)+αΔx+βΔy\Delta z = \Delta x\dfrac{\pd z}{\pd x}(x, y) + \Delta y\dfrac{\pd z}{\pd y}(x, y) + \alpha \Delta x + \beta \Delta y

ρ=(Δx)2+(Δy)2\rho = \sqrt{(\Delta x)^2 + (\Delta y)^2},则

0αΔx+βΔyραΔxρ+βΔyρα+β0(Δx0,Δy0)\begin{aligned} 0 &\le \dfrac{\left\lvert \alpha \Delta x + \beta \Delta y \right\rvert}{\rho}\\ &\le \left\lvert \alpha \right\rvert \dfrac{\left\lvert \Delta x \right\rvert}{\rho} + \left\lvert \beta \right\rvert \dfrac{\left\lvert \Delta y \right\rvert}{\rho}\\ &\le \left\lvert \alpha \right\rvert + \left\lvert \beta \right\rvert\\ &\to 0\quad (\Delta x \to 0,\, \Delta y \to 0) \end{aligned}

夹逼准则知

αΔx+βΔy=o(ρ)(ρ0+)\alpha \Delta x + \beta \Delta y = o(\rho)\quad (\rho \to 0^{+})

Δz=Δx ⁣z ⁣x(x,y)+Δy ⁣z ⁣y(x,y)+o(ρ)\Delta z = \Delta x\dfrac{\pd z}{\pd x}(x, y) + \Delta y\dfrac{\pd z}{\pd y}(x, y) + o(\rho)

从而 f(x,y)f(x, y) 在点 (x,y)(x, y) 可微。

若函数 z=f(x,y)z = f(x, y)(x,y)(x, y) 的某邻域内可偏导,且  ⁣z ⁣x\dfrac{\pd z}{\pd x} ⁣z ⁣y\dfrac{\pd z}{\pd y} 在点 (x,y)(x, y) 连续,则函数 z=f(x,y)z = f(x, y) 在点 (x,y)(x, y) 连续可微

z=f(x,y)z = f(x, y) 在开区域 GG 内每一点都连续可微,则称函数 z=f(x,y)z = f(x, y)GG 上连续可微

类似可导推不出导函数连续,可微并不代表导函数连续,因此连续可微的概念就与一元函数可导且导函数连续(即「连续可导」)类似。上面的证明和这里的「连续可微」定义一致,因为其实就是「连续可微」。

对于函数

z=f(x,y)={(x2+y2)sin1x2+y2,if (x,y)(0,0),0,if (x,y)=(0,0)z = f(x, y) = \begin{cases} (x^2 + y^2) \sin \dfrac{1}{x^2 + y^2}, & \text{if } (x, y) \neq (0, 0), \\ 0, & \text{if } (x, y) = (0, 0) \end{cases}

 ⁣z ⁣x(0,0)=limΔx0f(Δx,0)f(0,0)Δx=limΔx0Δxsin1Δx2=0\begin{aligned} \dfrac{\pd z}{\pd x}(0, 0) &= \lim\limits_{\Delta x \to 0} \dfrac{f(\Delta x, 0) - f(0, 0)}{\Delta x} \\ &= \lim\limits_{\Delta x \to 0} \Delta x \sin \dfrac{1}{\Delta x^2}\\ &= 0 \end{aligned}

同理可得  ⁣z ⁣y(0,0)=0\dfrac{\pd z}{\pd y}(0, 0) = 0

ω=Δz ⁣z ⁣xΔx ⁣z ⁣yΔy=Δz=f(Δx,Δy)f(0,0)=(Δx2+Δy2)sin1Δx2+Δy2\begin{aligned} \omega &= \Delta z - \dfrac{\pd z}{\pd x} \Delta x - \dfrac{\pd z}{\pd y} \Delta y \\ &= \Delta z \\ &= f(\Delta x, \Delta y) - f(0, 0) \\ &= (\Delta x^2 + \Delta y^2) \sin \dfrac{1}{\Delta x^2 + \Delta y^2} \end{aligned}

limΔx0Δy0ωρ=limΔx0Δy0Δx2+Δy2sin1Δx2+Δy2=0\begin{aligned} \lim\limits_{\substack{\Delta x \to 0\\ \Delta y \to 0}} \dfrac{\omega}{\rho} &= \lim\limits_{\substack{\Delta x \to 0\\ \Delta y \to 0}} \sqrt{\Delta x^2 + \Delta y^2} \sin \dfrac{1}{\Delta x^2 + \Delta y^2} \\ &= 0 \end{aligned}

ω\omegaρ\rho 的高阶无穷小,Δz\Delta z 可以表示成 AΔx+BΔy+o(ρ)A \Delta x + B \Delta y + o(\rho) 的形式,因此 f(x)f(x)(0,0)(0, 0) 处可微。

f(x)f(x)xx 的偏导数

 ⁣z ⁣x=2x(sin1x2+y21x2+y2cos1x2+y2)\begin{aligned} \dfrac{\pd z}{\pd x} &= 2 x \left(\sin \frac{1}{x^{2}+y^{2}}-\frac{1}{x^{2}+y^{2}} \cos \frac{1}{x^{2}+y^{2}}\right) \end{aligned}

(0,0)(0, 0) 的邻域内无界,从而  ⁣z ⁣x\dfrac{\pd z}{\pd x}(0,0)(0, 0) 处不连续。

由此说明了可微无法推出导函数连续。

二元函数的结论,也可以推广到多元函数,如函数 z=f(x1,x2,,xn)z = f(x_1, x_2, \cdots, x_n) 的全微分

 ⁣dz=i=1n ⁣z ⁣xi ⁣dxi\d z = \sum_{i=1}^{n} \dfrac{\pd z}{\pd x_i}\d x_i

不再赘述。

目前看来,多元函数有四个性质:可偏导、连续、可微、连续可微。它们大致有下面的关系:

  • 可偏导不一定连续,连续也不一定可偏导
  • 可微一定可偏导,可微一定连续
  • 可偏导、连续不一定可微
  • 可微不一定连续可微,连续可微一定可微

设函数 z=f(x,y)z = f(x, y) 在点 (x,y)(x, y) 处的所有 nn 阶偏导数都存在且连续,则函数 z=f(x,y)z = f(x, y) 在点 (x,y)(x, y)nn 阶可微(即函数 f(x,y)f(x, y) nn 阶连续可微),且有

 ⁣dnz=( ⁣dx ⁣ ⁣x+ ⁣dy ⁣ ⁣y)nf(x,y)\d^n z = \left( \d x \dfrac{\pd }{\pd x} + \d y \dfrac{\pd }{\pd y} \right)^n f(x, y)

或者展开有

 ⁣dnz=k=0n(nk) ⁣nf ⁣xk ⁣ynk(x,y) ⁣dxk ⁣dynk\d^n z = \sum_{k=0}^n \dbinom{n}{k} \dfrac{\pd^n f}{\pd x^k \pd y^{n - k}} (x, y) \d x^k \d y^{n - k}