如何理解数学中的梯度(图文版)
马同学图解数学
2024年08月20日 16:19
收录于文集
共13篇

梯度(tī dù)中文意思为坡度。在数学中,它也是坡度,不过是最陡的那个坡的坡度。严格说:梯度是一个向量,其方向为变化最快的那个方向,其大小就是那个方向的变化率。

1 梯度的表达式

由于在数学中,表示描述一个方向上的变化率,用的是方向导数。那么根据前面所说,梯度的值就应该是所有方向导数中的最大者。

定理 . 如果函数f(x%2Cy) 在(x_0%2Cy_0) 点可微分,那么函数f(x%2Cy) 在该点沿任意方向%5Cboldsymbol%7Bu%7D 的方向导数存在。设%5Cboldsymbol%7Bu%7D 的单位方向向量%5Cboldsymbol%7Be%7D_%7B%5Cboldsymbol%7Bu%7D%7D%3D%5Cbegin%7Bpmatrix%7D%5Ccos%5Calpha%5C%5C%5Ccos%5Cbeta%5Cend%7Bpmatrix%7D ,则有:

%5Cleft.%5Cfrac%7B%5Cpartial%2Cf%7D%7B%5Cpartial%2C%5Cboldsymbol%7Bu%7D%7D%5Cright%7C_%7B%5Cbegin%7Baligned%7Dx%3Dx_0%5C%5Cy%3Dy_0%5Cend%7Baligned%7D%7D%3Df_x(x_0%2Cy_0)%5Ccos%5Calpha%2Bf_y(x_0%2Cy_0)%5Ccos%5Cbeta

证明 . 因为函数f(x%2Cy) 在(x_0%2Cy_0) 点可微分,所以有:

f(x_0%2B%5CDelta%2Cx%2C%2Cy_0%2B%5CDelta%2Cy)-f(x_0%2C%2Cy_0)%3Df_x(x_0%2Cy_0)%5CDelta%2Cx%2Bf_y(x_0%2Cy_0)%5CDelta%2Cy%2Bo(%5Csqrt%7B(%5CDelta%2Cx)%5E2%2B(%5CDelta%2Cy)%5E2%7D)

(x_0%2B%5CDelta%2Cx%2Cy_0%2B%5CDelta%2Cy) 在以(x_0%2Cy_0) 点为起点、方向为%5Cboldsymbol%7Bu%7D 的射线l 上时,有:

%5CDelta%2Cx%3Dt%5Ccos%5Calpha%2C%5Cquad%5CDelta%2Cy%3Dt%5Ccos%5Cbeta%2C%5Cquad%5Csqrt%7B(%5CDelta%2Cx)%5E2%2B(%5CDelta%2Cy)%5E2%7D%3Dt

结合上方向导数的定义,所以:

%0A%5Cbegin%7Baligned%7D%0A%2C%2C%2C%2C%5Cleft.%5Cfrac%7B%5Cpartial%2Cf%7D%7B%5Cpartial%2C%5Cboldsymbol%7Bu%7D%7D%5Cright%7C_%7B%5Cbegin%7Baligned%7Dx%3Dx_0%5C%5Cy%3Dy_0%5Cend%7Baligned%7D%7D%0A%2C%2C%2C%2C%2C%2C%2C%2C%26%3D%5Clim_%7Bt%5Cto%2C0%5E%2B%7D%5Cfrac%7Bf(x_0%2Bt%5Ccos%5Calpha%2C%2Cy_0%2Bt%5Ccos%5Cbeta)-f(x_0%2Cy_0)%7D%7Bt%7D%5C%5C%0A%2C%2C%2C%2C%2C%2C%2C%2C%26%3D%5Clim_%7Bt%5Cto%2C0%5E%2B%7D%5Cfrac%7Bf(x_0%2B%5CDelta%2Cx%2C%2Cy_0%2B%5CDelta%2Cy)-f(x_0%2Cy_0)%7D%7Bt%7D%5C%5C%0A%2C%2C%2C%2C%2C%2C%2C%2C%26%3D%5Clim_%7Bt%5Cto%2C0%5E%2B%7D%5Cfrac%7Bf_x(x_0%2Cy_0)%5CDelta%2Cx%2Bf_y(x_0%2Cy_0)%5CDelta%2Cy%2Bo(%5Csqrt%7B(%5CDelta%2Cx)%5E2%2B(%5CDelta%2Cy)%5E2%7D)%7D%7Bt%7D%5C%5C%0A%2C%2C%2C%2C%2C%2C%2C%2C%26%3D%5Clim_%7Bt%5Cto%2C0%5E%2B%7D%5Cfrac%7Bf_x(x_0%2Cy_0)%5Ccdot%2Ct%5Ccos%5Calpha%2Bf_y(x_0%2Cy_0)%5Ccdot%2Ct%5Ccos%5Cbeta%2Bo(t)%7D%7Bt%7D%5C%5C%0A%2C%2C%2C%2C%2C%2C%2C%2C%26%3Df_x(x_0%2Cy_0)%5Ccos%5Calpha%2Bf_y(x_0%2Cy_0)%5Ccos%5Cbeta%0A%5Cend%7Baligned%7D%0A

上面的定理说明,在可微的情况下,我们可以得到方向导数是偏导数的线性组合。其中下标位置的x_0%2Cy_0 是这个点的位置。而%5Calpha 和%5Cbeta 分别为两个偏导数与方向向量%5Cboldsymbol%7Bu%7D 的夹角。

可微分时方向导数的计算方法可改写如下:

%0A%5Cbegin%7Baligned%7D%0A%2C%2C%2C%2C%5Cleft.%5Cfrac%7B%5Cpartial%2Cf%7D%7B%5Cpartial%2C%5Cboldsymbol%7Bu%7D%7D%5Cright%7C_%7B%5Cbegin%7Baligned%7Dx%3Dx_0%5C%5Cy%3Dy_0%5Cend%7Baligned%7D%7D%0A%2C%2C%2C%2C%2C%2C%2C%2C%26%3Df_x(x_0%2Cy_0)%5Ccos%5Calpha%2Bf_y(x_0%2Cy_0)%5Ccos%5Cbeta%3D%5Cunderbrace%7B%5Cbegin%7Bpmatrix%7Df_x(x_0%2Cy_0)%5C%5Cf_y(x_0%2Cy_0)%5Cend%7Bpmatrix%7D%7D_%7B%5Cboldsymbol%7Bv%7D%7D%5Ccdot%5Cunderbrace%7B%5Cbegin%7Bpmatrix%7D%5Ccos%5Calpha%5C%5C%5Ccos%5Cbeta%5Cend%7Bpmatrix%7D%7D_%7B%5Cboldsymbol%7Be%7D_u%7D%0A%5Cend%7Baligned%7D

改写后,可以看出方向导数是%5Cboldsymbol%7Bu%7D 在%5Cboldsymbol%7Be%7D_u 上的投影。则,当两者不断靠拢时,投影就不会不断变大。而当%5Cboldsymbol%7Be%7D_u 与%5Cboldsymbol%7Bv%7D 同向时,所得投影获得最大值。

在我们已经知道了梯度的定义后,就会发现,函数在%5Cboldsymbol%7Bu%7D 方向上的方向导数。它就应该等于,梯度在其方向向量上的投影。

%0A%5Cbegin%7Baligned%7D%0A%2C%2C%2C%2C%5Cleft.%5Cfrac%7B%5Cpartial%2Cf%7D%7B%5Cpartial%2C%5Cboldsymbol%7Bu%7D%7D%5Cright%7C_%7B%5Cbegin%7Baligned%7Dx%3Dx_0%5C%5Cy%3Dy_0%5Cend%7Baligned%7D%7D%0A%2C%2C%2C%2C%2C%2C%2C%2C%26%3Df_x(x_0%2Cy_0)%5Ccos%5Calpha%2Bf_y(x_0%2Cy_0)%5Ccos%5Cbeta%3D%5Cbegin%7Bpmatrix%7Df_x(x_0%2Cy_0)%5C%5Cf_y(x_0%2Cy_0)%5Cend%7Bpmatrix%7D%5Ccdot%5Cbegin%7Bpmatrix%7D%5Ccos%5Calpha%5C%5C%5Ccos%5Cbeta%5Cend%7Bpmatrix%7D%5C%5C%0A%2C%2C%2C%2C%2C%2C%2C%2C%26%3D%5Cnabla%2Cf(x_0%2Cy_0)%5Ccdot%5Cboldsymbol%7Be%7D_%7B%5Cboldsymbol%7Bu%7D%7D%3D%5C%7C%5Cnabla%2Cf(x_0%2Cy_0)%5C%7C%5Ccos%5Ctheta%0A%5Cend%7Baligned%7D

而这里的%5Ctheta ,反映在图像上。就是梯度与方向向量%5Cboldsymbol%7Be%7D_u 的夹角。这样我们就知道:

  •  时,即方向 与梯度 相同时,或者说沿着梯度方向时,方向导数取得最大值:

  •  时,即方向 与梯度 相反时,或者说逆着梯度方向时,方向导数取得最小值:

%5Cleft.%5Cfrac%7B%5Cpartial%2Cf%7D%7B%5Cpartial%2C%5Cboldsymbol%7Bu%7D%7D%5Cright%7C_%7B%5Cbegin%7Baligned%7Dx%3Dx_0%5C%5Cy%3Dy_0%5Cend%7Baligned%7D%7D%3D%5C%7C%5Cnabla%2Cf(x_0%2Cy_0)%5C%7C%5Ccos%5Cpi%3D-%5C%7C%5Cnabla%2Cf(x_0%2Cy_0)%5C%7C

  •  时,即方向 与梯度 正交时,方向导数为 :

%5Cleft.%5Cfrac%7B%5Cpartial%2Cf%7D%7B%5Cpartial%2C%5Cboldsymbol%7Bu%7D%7D%5Cright%7C_%7B%5Cbegin%7Baligned%7Dx%3Dx_0%5C%5Cy%3Dy_0%5Cend%7Baligned%7D%7D%3D%5C%7C%5Cnabla%2Cf(x_0%2Cy_0)%5C%7C%5Ccos%5Cfrac%7B%5Cpi%7D%7B2%7D%3D0

3 用梯度表示所有的方向导数

根据这个理论,我们可以做出某二元函数在(0,0)点处,所有的方向导数。其中蓝色向量的模长,表示其所在方向的方向导数,红色向量,为函数在此点处的梯度。


少侠我看你骨骼惊奇,天赋异禀,定是学数学的好苗子,来一套马同学图解数学秘籍吧!包你一学就会,从此数学之路所向披靡!