这个很容易理解,跟一元函数一样。
例如 f ( x , y ) = ∣ x ∣ f(x,y)=|x| f(x,y)=∣x∣,在 ( 0 , 0 ) (0,0) (0,0)连续,但 f x ( 0 , 0 ) = d ∣ x ∣ d x f_x(0,0)=\frac{\text{d}|x|}{\text{d}x} fx(0,0)=dxd∣x∣不存在。
再例如, f ( x , y ) = x 2 + y 2 f(x,y)=\sqrt{x^2+y^2} f(x,y)=x2+y2,其在 ( 0 , 0 ) (0,0) (0,0)点显然连续,但 f x ( 0 , 0 ) = lim x → 0 ∣ x ∣ x f_x(0,0)=\lim_{x\to0}\frac{|x|}{x} fx(0,0)=x→0limx∣x∣不存在, f y ( 0 , 0 ) f_y(0,0) fy(0,0)同理也不存在。
用Geogebra画图可以看出这个函数的图像是锥形,在 ( 0 , 0 ) (0,0) (0,0)点是尖的:
这个性质跟一元函数有很大差异。对于二元函数,偏导数存在是很弱的条件,甚至连极限都有可能不存在。
例子: f ( x , y ) = { x y x 2 + y 2 , x 2 + y 2 ≠ 0 0 , x 2 + y 2 = 0 f(x,y)=\begin{cases}\frac{xy}{x^2+y^2},&x^2+y^2\ne0\\0,&x^2+y^2=0\end{cases} f(x,y)={x2+y2xy,0,x2+y2=0x2+y2=0它在 ( 0 , 0 ) (0,0) (0,0)点的两个偏导数都存在: f x ( 0 , 0 ) = f y ( 0 , 0 ) = 0 f_x(0,0)=f_y(0,0)=0 fx(0,0)=fy(0,0)=0但是它在 ( 0 , 0 ) (0,0) (0,0)点的极限不存在,以 y = k x y=kx y=kx的路径逼近 ( 0 , 0 ) (0,0) (0,0)得 lim x → 0 , y = k x x y x 2 + y 2 = k 1 + k 2 \lim_{x\to0,y=kx}\frac{xy}{x^2+y^2}=\frac{k}{1+k^2} x→0,y=kxlimx2+y2xy=1+k2k随着 k k k的变化而变化,所以 lim ( x , y ) → ( 0 , 0 ) f ( x , y ) \lim_{(x,y)\to(0,0)}f(x,y) (x,y)→(0,0)limf(x,y)不存在。
画图看出这个函数在 ( 0 , 0 ) (0,0) (0,0)点呈现一个很奇怪的样子:
定理1(可微的必要条件) 设函数 z = f ( x , y ) z=f(x,y) z=f(x,y)在 ( x 0 , y 0 ) (x_0,y_0) (x0,y0)处可微,则
(1) f f f在 ( x 0 , y 0 ) (x_0,y_0) (x0,y0)处连续;
(2) f f f在 ( x 0 , y 0 ) (x_0,y_0) (x0,y0)处的两个偏导数都存在,且有 d f ( x 0 , y 0 ) = f x ( x 0 , y 0 ) d x + f y ( x 0 , y 0 ) d y \text{d}f(x_0,y_0)=f_x(x_0,y_0)\text{d}x+f_y(x_0,y_0)\text{d}y df(x0,y0)=fx(x0,y0)dx+fy(x0,y0)dy。
证明:
(1) 当 f f f在 ( x 0 , y 0 ) (x_0,y_0) (x0,y0)处可微时,存在常数 a 1 , a 2 a_1,a_2 a1,a2使得 Δ z = a 1 Δ x + a 2 Δ y + o ( ρ ) \Delta z=a_1\Delta x+a_2\Delta y+o(\rho) Δz=a1Δx+a2Δy+o(ρ),其中 ρ = ( Δ x ) 2 + ( Δ y ) 2 \rho=\sqrt{(\Delta x)^2+(\Delta y)^2} ρ=(Δx)2+(Δy)2。令 ρ → 0 \rho\to0 ρ→0,即 Δ x → 0 \Delta x\to0 Δx→0且 Δ y → 0 \Delta y\to0 Δy→0,得 lim ρ → 0 Δ z = 0 \lim_{\rho\to0}\Delta z=0 ρ→0limΔz=0或 lim Δ x → 0 , Δ y → 0 f ( x 0 + Δ x , y 0 + Δ y ) = lim Δ x → 0 , Δ y → 0 [ f ( x 0 , y 0 ) + Δ z ] = f ( x 0 , y 0 ) \lim_{\Delta x\to0,\Delta y\to0}f(x_0+\Delta x,y_0+\Delta y)=\lim_{\Delta x\to0,\Delta y\to0}[f(x_0,y_0)+\Delta z]=f(x_0,y_0) Δx→0,Δy→0limf(x0+Δx,y0+Δy)=Δx→0,Δy→0lim[f(x0,y0)+Δz]=f(x0,y0)因此 f f f在 ( x 0 , y 0 ) (x_0,y_0) (x0,y0)处连续。
(2) 由可微的定义, f f f满足 f ( x 0 + Δ x , y 0 + Δ y ) − f ( x 0 , y 0 ) = a 1 Δ x + a 2 Δ y + o ( ρ ) f(x_0+\Delta x,y_0+\Delta y)-f(x_0,y_0)=a_1\Delta x+a_2\Delta y+o(\rho) f(x0+Δx,y0+Δy)−f(x0,y0)=a1Δx+a2Δy+o(ρ)取 Δ y = 0 \Delta y=0 Δy=0,则有 ρ = ∣ Δ x ∣ \rho=|\Delta x| ρ=∣Δx∣,上式变为 f ( x 0 + Δ x , y 0 ) − f ( x 0 , y 0 ) = a 1 + o ( ∣ Δ x ∣ ) f(x_0+\Delta x,y_0)-f(x_0,y_0)=a_1+o(|\Delta x|) f(x0+Δx,y0)−f(x0,y0)=a1+o(∣Δx∣)两边除以 Δ x \Delta x Δx并取极限得 lim Δ x → 0 f ( x 0 + Δ x , y 0 ) − f ( x 0 , y 0 ) Δ x = lim Δ x → 0 [ a 1 + o ( ∣ Δ x ∣ ) Δ x ] = a 1 \lim_{\Delta x\to0}\frac{f(x_0+\Delta x,y_0)-f(x_0,y_0)}{\Delta x}=\lim_{\Delta x\to0}\left[a_1+\frac{o(|\Delta x|)}{\Delta x}\right]=a_1 Δx→0limΔxf(x0+Δx,y0)−f(x0,y0)=Δx→0lim[a1+Δxo(∣Δx∣)]=a1即 f x ( x 0 , y 0 ) = a 1 f_x(x_0,y_0)=a_1 fx(x0,y0)=a1。
同理,取 Δ x = 0 \Delta x=0 Δx=0得 f y ( x 0 , y 0 ) = a 2 f_y(x_0,y_0)=a_2 fy(x0,y0)=a2。∎
从这里我们可以看出,可微是很强得条件,远比偏导数存在要强。
然而,这个条件仅仅是必要条件。我们举一个例子 f ( x , y ) = { x y x 2 + y 2 , x 2 + y 2 ≠ 0 0 , x 2 + y 2 = 0 f(x,y)=\begin{cases}\frac{xy}{\sqrt{x^2+y^2}},&x^2+y^2\ne0\\0,&x^2+y^2=0\end{cases} f(x,y)={x2+y2xy,0,x2+y2=0x2+y2=0它在 ( 0 , 0 ) (0,0) (0,0)点连续,因为 lim ( x , y ) → ( 0 , 0 ) ∣ f ( x , y ) − f ( 0 , 0 ) ∣ = lim ( x , y ) → ( 0 , 0 ) ∣ f ( x , y ) ∣ ≤ lim ( x , y ) → ( 0 , 0 ) x 2 + y 2 2 x 2 + y 2 = 0 \lim_{(x,y)\to(0,0)}|f(x,y)-f(0,0)|=\lim_{(x,y)\to(0,0)}|f(x,y)|\le\lim_{(x,y)\to(0,0)}\frac{x^2+y^2}{2\sqrt{x^2+y^2}}=0 (x,y)→(0,0)lim∣f(x,y)−f(0,0)∣=(x,y)→(0,0)lim∣f(x,y)∣≤(x,y)→(0,0)lim2x2+y2x2+y2=0两个偏导数也存在: f x ( 0 , 0 ) = f y ( 0 , 0 ) = 0 f_x(0,0)=f_y(0,0)=0 fx(0,0)=fy(0,0)=0但不可微。因为如果可微,那么 Δ f − f x ( 0 , 0 ) Δ x − f y ( 0 , 0 ) Δ y = o ( ρ ) \Delta f-f_x(0,0)\Delta x-f_y(0,0)\Delta y=o(\rho) Δf−fx(0,0)Δx−fy(0,0)Δy=o(ρ)。然而 lim ( x , y ) → ( 0 , 0 ) Δ f ρ = lim ( x , y ) → ( 0 , 0 ) x y x 2 + y 2 \lim_{(x,y)\to(0,0)}\frac{\Delta f}{\rho}=\lim_{(x,y)\to(0,0)}\frac{xy}{x^2+y^2} (x,y)→(0,0)limρΔf=(x,y)→(0,0)limx2+y2xy不存在。也就是说,满足定理1的条件不一定可微。
其函数图像如下:
定理2(可微的充分条件) 设函数 z = f ( x , y ) z=f(x,y) z=f(x,y)在点 ( x 0 , y 0 ) (x_0,y_0) (x0,y0)得的某个邻域内有定义,若 f ( x , y ) f(x,y) f(x,y)的两个偏导数 f x ( x , y ) f_x(x,y) fx(x,y)和 f y ( x , y ) f_y(x,y) fy(x,y)均在点 ( x 0 , y 0 ) (x_0,y_0) (x0,y0)处连续,则该函数在点 ( x 0 , y 0 ) (x_0,y_0) (x0,y0)处可微。
证明: Δ z = f ( x 0 + Δ x , y 0 + Δ y ) − f ( x 0 , y 0 ) = [ f ( x 0 + Δ x , y 0 + Δ y ) − f ( x 0 , y 0 + Δ y ) ] + [ f ( x 0 , y 0 + Δ y ) − f ( x 0 , y 0 ) ] \begin{aligned}\Delta z&=f(x_0+\Delta x,y_0+\Delta y)-f(x_0,y_0)\\&=[f(x_0+\Delta x,y_0+\Delta y)-f(x_0,y_0+\Delta y)]+[f(x_0,y_0+\Delta y)-f(x_0,y_0)]\end{aligned} Δz=f(x0+Δx,y0+Δy)−f(x0,y0)=[f(x0+Δx,y0+Δy)−f(x0,y0+Δy)]+[f(x0,y0+Δy)−f(x0,y0)]右边的每一项都是一元函数的改变量,故可以采用拉格朗日中值定理( f ( b ) − f ( a ) = f ′ ( ξ ) ( b − a ) f(b)-f(a)=f'(\xi)(b-a) f(b)−f(a)=f′(ξ)(b−a)),即存在 θ 1 , θ 2 ∈ ( 0 , 1 ) \theta_1,\theta_2\in(0,1) θ1,θ2∈(0,1)使得 Δ z = f x ( x 0 + θ 1 Δ x , y 0 + Δ y ) Δ x + f y ( x 0 , y 0 + θ 2 Δ y ) Δ y \Delta z=f_x(x_0+\theta_1\Delta x,y_0+\Delta y)\Delta x+f_y(x_0,y_0+\theta_2\Delta y)\Delta y Δz=fx(x0+θ1Δx,y0+Δy)Δx+fy(x0,y0+θ2Δy)Δy由于 f x ( x , y ) f_x(x,y) fx(x,y)在 ( x 0 , y 0 ) (x_0,y_0) (x0,y0)连续,取极限 ρ = ( Δ x ) 2 + ( Δ y ) 2 → 0 \rho=\sqrt{(\Delta x)^2+(\Delta y)^2}\to0 ρ=(Δx)2+(Δy)2→0得 lim ρ → 0 f x ( x 0 + θ 1 Δ x , y 0 + Δ y ) = f x ( x 0 , y 0 ) \lim_{\rho\to0}f_x(x_0+\theta_1\Delta x,y_0+\Delta y)=f_x(x_0,y_0) ρ→0limfx(x0+θ1Δx,y0+Δy)=fx(x0,y0)因此有 f x ( x 0 + θ 1 Δ x , y 0 + Δ y ) = f x ( x 0 , y 0 ) + α 1 ( ρ ) f_x(x_0+\theta_1\Delta x,y_0+\Delta y)=f_x(x_0,y_0)+\alpha_1(\rho) fx(x0+θ1Δx,y0+Δy)=fx(x0,y0)+α1(ρ)同理有 f y ( x 0 , y 0 + θ 2 Δ y ) = f ( x 0 , y 0 ) + α 2 ( ρ ) f_y(x_0,y_0+\theta_2\Delta y)=f(x_0,y_0)+\alpha_2(\rho) fy(x0,y0+θ2Δy)=f(x0,y0)+α2(ρ)其中 α 1 , 2 ( ρ ) \alpha_{1,2}( \rho) α1,2(ρ)是 ρ \rho ρ的高阶无穷小。整理得 Δ z = f ( x 0 + Δ x , y 0 + Δ y ) − f ( x 0 , y 0 ) = [ f ( x 0 + Δ x , y 0 + Δ y ) − f ( x 0 , y 0 + Δ y ) ] + [ f ( x 0 , y 0 + Δ y ) − f ( x 0 , y 0 ) ] = [ f x ( x 0 , y 0 ) + α 1 ( ρ ) ] Δ x + [ f ( x 0 , y 0 ) + α 2 ( ρ ) ] Δ y = f x ( x 0 , y 0 ) Δ x + f y ( x 0 , y 0 ) Δ y + α 1 ( ρ ) Δ x + α 2 ( ρ ) Δ y \begin{aligned}\Delta z&=f(x_0+\Delta x,y_0+\Delta y)-f(x_0,y_0)\\&=[f(x_0+\Delta x,y_0+\Delta y)-f(x_0,y_0+\Delta y)]+[f(x_0,y_0+\Delta y)-f(x_0,y_0)]\\&=[f_x(x_0,y_0)+\alpha_1(\rho)]\Delta x+[f(x_0,y_0)+\alpha_2(\rho)]\Delta y\\&=f_x(x_0,y_0)\Delta x+f_y(x_0,y_0)\Delta y+\alpha_1(\rho)\Delta x+\alpha_2(\rho)\Delta y\end{aligned} Δz=f(x0+Δx,y0+Δy)−f(x0,y0)=[f(x0+Δx,y0+Δy)−f(x0,y0+Δy)]+[f(x0,y0+Δy)−f(x0,y0)]=[fx(x0,y0)+α1(ρ)]Δx+[f(x0,y0)+α2(ρ)]Δy=fx(x0,y0)Δx+fy(x0,y0)Δy+α1(ρ)Δx+α2(ρ)Δy只需证明后面两项是 ρ \rho ρ的高阶无穷小。而 Δ x ≤ ρ \Delta x\le\rho Δx≤ρ, Δ y ≤ ρ \Delta y\le\rho Δy≤ρ,所以 ∣ α 1 ( ρ ) Δ x + α 2 ( ρ ) Δ y ∣ ≤ ∣ α 1 ( ρ ) + α 2 ( ρ ) ∣ ρ |\alpha_1(\rho)\Delta x+\alpha_2(\rho)\Delta y|\le|\alpha_1(\rho)+\alpha_2(\rho)|\rho ∣α1(ρ)Δx+α2(ρ)Δy∣≤∣α1(ρ)+α2(ρ)∣ρ, lim ρ → 0 ∣ α 1 ( ρ ) Δ x + α 2 ( ρ ) Δ y ρ ∣ = lim ρ → 0 α 1 ( ρ ) + α 2 ( ρ ) = o ( ρ ) \lim_{\rho\to0}\left|\frac{\alpha_1(\rho)\Delta x+\alpha_2(\rho)\Delta y}{\rho}\right|=\lim_{\rho\to0}\alpha_1(\rho)+\alpha_2(\rho)=o(\rho) ρ→0lim∣∣∣∣ρα1(ρ)Δx+α2(ρ)Δy∣∣∣∣=ρ→0limα1(ρ)+α2(ρ)=o(ρ)于是 α 1 ( ρ ) Δ x + α 2 ( ρ ) Δ y \alpha_1(\rho)\Delta x+\alpha_2(\rho)\Delta y α1(ρ)Δx+α2(ρ)Δy是 ρ \rho ρ的高阶无穷小。因此有 Δ z = f x ( x 0 , y 0 ) Δ x + f y ( x 0 , y 0 ) Δ y + o ( ρ ) \Delta z=f_x(x_0,y_0)\Delta x+f_y(x_0,y_0)\Delta y+o(\rho) Δz=fx(x0,y0)Δx+fy(x0,y0)Δy+o(ρ)即 f f f在 ( x 0 , y 0 ) (x_0,y_0) (x0,y0)处可微。∎
注意:这只是充分条件。有些函数,例如 f ( x , y ) = { ( x 2 + y 2 ) sin 1 x 2 + y 2 , x 2 + y 2 ≠ 0 0 , x 2 + y 2 = 0 f(x,y)=\begin{cases}(x^2+y^2)\sin{\frac{1}{x^2+y^2}},&x^2+y^2\ne0\\0,&x^2+y^2=0\end{cases} f(x,y)={(x2+y2)sinx2+y21,0,x2+y2=0x2+y2=0它在 ( 0 , 0 ) (0,0) (0,0)处可微,但 f x ( x , y ) f_x(x,y) fx(x,y)与 f y ( x , y ) f_y(x,y) fy(x,y)在 ( 0 , 0 ) (0,0) (0,0)处间断。
另外, f x ( x , y ) , f y ( x , y ) f_x(x,y),f_y(x,y) fx(x,y),fy(x,y)是二元函数,它们连续是指满足二元函数连续的条件,而不仅仅是在 x x x方向或在 y y y方向上连续。
这是定理1和定理2结合起来后一个很显然的推论。
定理3 若 f ( x , y ) f(x,y) f(x,y)在点 ( x 0 , y 0 ) (x_0,y_0) (x0,y0)可微,则函数 f f f在点 ( x 0 , y 0 ) (x_0,y_0) (x0,y0)沿任意 l \bm l l方向的方向导数均存在,且 ∂ f ∂ l ∣ x 0 , y 0 = f x ( x 0 , y 0 ) cos α + f y ( x 0 , y 0 ) cos β \left.\frac{\partial f}{\partial\bm l}\right|_{x_0,y_0}=f_x(x_0,y_0)\cos\alpha+f_y(x_0,y_0)\cos\beta ∂l∂f∣∣∣∣x0,y0=fx(x0,y0)cosα+fy(x0,y0)cosβ其中 l \bm l l方向上的单位向量是 e l = ( cos α , cos β ) \bm e_l=(\cos\alpha,\cos\beta) el=(cosα,cosβ)。
证明:由定理1,当 ( x , y ) → ( 0 , 0 ) (x,y)\to(0,0) (x,y)→(0,0)时,有 f ( x , y ) − f ( x 0 , y 0 ) = f x ( x 0 , y 0 ) Δ x + f y ( x 0 , y 0 ) Δ y + o ( ρ ) f(x,y)-f(x_0,y_0)=f_x(x_0,y_0)\Delta x+f_y(x_0,y_0)\Delta y+o(\rho) f(x,y)−f(x0,y0)=fx(x0,y0)Δx+fy(x0,y0)Δy+o(ρ)令 ( x , y ) = ( x 0 , y 0 ) + t e l = ( x 0 , y 0 ) + ( t cos α , t cos β ) (x,y)=(x_0,y_0)+t\bm e_l=(x_0,y_0)+(t\cos\alpha,t\cos\beta) (x,y)=(x0,y0)+tel=(x0,y0)+(tcosα,tcosβ),即 Δ x = t cos α , Δ y = t cos β , ∣ t ∣ = ( Δ x ) 2 + ( Δ y ) 2 \Delta x=t\cos\alpha,\Delta y=t\cos\beta,|t|=\sqrt{(\Delta x)^2+(\Delta y)^2} Δx=tcosα,Δy=tcosβ,∣t∣=(Δx)2+(Δy)2,可得 f ( x 0 , y 0 ) = f x ( x 0 , y 0 ) t cos α + f y ( x 0 , y 0 ) t cos β + o ( ρ ) f(x_0,y_0)=f_x(x_0,y_0)t\cos\alpha+f_y(x_0,y_0)t\cos\beta+o(\rho) f(x0,y0)=fx(x0,y0)tcosα+fy(x0,y0)tcosβ+o(ρ)由方向导数的定义有 ∂ f ∂ l ∣ x 0 , y 0 = lim t → 0 f ( x 0 + t cos α , y 0 + t cos β ) − f ( x 0 , y 0 ) t = lim t → 0 f x ( x 0 , y 0 ) t cos α + f y ( x 0 , y 0 ) t cos β + o ( ∣ t ∣ ) t = f x ( x 0 , y 0 ) cos α + f y ( x 0 , y 0 ) cos β \begin{aligned}\left.\frac{\partial f}{\partial\bm l}\right|_{x_0,y_0}&=\lim_{t\to0}\frac{f(x_0+t\cos\alpha,y_0+t\cos\beta)-f(x_0,y_0)}{t}\\&=\lim_{t\to0}\frac{f_x(x_0,y_0)t\cos\alpha+f_y(x_0,y_0)t\cos\beta+o(|t|)}{t}\\&=f_x(x_0,y_0)\cos\alpha+f_y(x_0,y_0)\cos\beta\end{aligned} ∂l∂f∣∣∣∣x0,y0=t→0limtf(x0+tcosα,y0+tcosβ)−f(x0,y0)=t→0limtfx(x0,y0)tcosα+fy(x0,y0)tcosβ+o(∣t∣)=fx(x0,y0)cosα+fy(x0,y0)cosβ证毕。∎
综合以上所有讨论,我们将各个条件之间的关系理成下面这张图:
可以看出偏导数连续是最强的条件,可微是很强的条件,(任意方向)偏导数存在是很弱的条件。