[数学]变分初涉

变分初涉

变分定义

设函数 $y=f(x)$ 过两个端点 $(x_0,y_0)$ 和 $(x_1,y_1)$, 任取过 $(x_0,0)$ 和 $(x_1,0)$ 的光滑可导函数 $\eta (x)$, 设无穷小为 $ϵ$, 则定义函数 $f$ 的变分为 $$\delta y=ϵ\eta$$

相应地, 定义 $$\begin{gathered} \delta y^{\prime }=(\delta y{)}^{\prime }=ϵ{\eta }^{\prime } \\ \delta y^{\prime \prime }=(\delta y{)}^{\prime \prime }=ϵ{\eta }^{\prime \prime } \end{gathered}$$ 等. 注意, 一般在同一个上下文中, 所有函数的变分共用一个 $\eta$ 函数及其导数, 故有 $$\frac{\text{d}(\delta y)}{\text{d}x}=\delta \left(\frac{\text{d}y}{\text{d}x}\right)$$

以上定义和自变量 $x$ 的微分 $\text{d}x$ 一样, 单独取出来考虑并无实际意义.
现在在此基础上, 考虑关于函数 $f$ 的泛函 $F(x,y,y^{\prime })$, 定义泛函的变分为 $$\delta F=F(x,y+\delta y,y^{\prime }+\delta y^{\prime })-F(y,y^{\prime })$$

得到 $$\delta F=\frac{\partial F}{\partial y}\delta y+\frac{\partial F}{\partial y^{\prime }}\delta y^{\prime }$$

一般地, 泛函 $F(x,y,y^{\prime },y^{\prime \prime },\cdots )$ 亦有 $$\delta F=\sum _{i=0}^{\infty }\frac{\partial F}{\partial y^{(i)}}\delta y^{(i)}$$

例如
$$\delta (y{y}^{\prime })=y^{\prime }\delta y+y\delta y^{\prime }$$

泛函极值

变分学的泛函指的是将函数映射为值的映射.

给予两个定点 $(x_0,y_0)$ 和 $(x_1,y_1)$, 过两点作曲线 $y=f(x)$, 满足 $f(x_0)=y_0$ 和 $f(x_1)=y_1$.
考虑定积分定义的泛函 $$I(f)={\int }_{x_0}^{x_1}F(x,y,y^{\prime })\text{d}x$$ 希望求得一个 $y$ 使得 $I$ 取极值.

给函数 $F$ 做变分, 得到 $$\delta F=\frac{\partial F}{\partial y}\delta y+\frac{\partial F}{\partial y^{\prime }}\delta y^{\prime }$$

其中变分 $\delta y$ 满足 ${\delta y|}_{x=x_0}=0$ 和 ${\delta y|}_{x=x_1}=0$.

$I$ 亦发生微小变化
$$\delta I=\delta {\int }_{x_0}^{x_1}F\text{d}x={\int }_{x_0}^{x_1}\left(\frac{\partial F}{\partial y}\delta y+\frac{\partial F}{\partial y^{\prime }}\delta y^{\prime }\right)\text{d}x$$

继而
$$\begin{array}{rl}\delta I=& {\int }_{x_0}^{x_1}\frac{\partial F}{\partial y}\delta y\text{d}x+{\int }_{x_0}^{x_1}\frac{\partial F}{\partial y^{\prime }}\delta \frac{\text{d}y}{\text{d}x}\text{d}x\\ =& {\int }_{x_0}^{x_1}\frac{\partial F}{\partial y}\delta y\text{d}x+{\int }_{x_0}^{x_1}\frac{\partial F}{\partial y^{\prime }}\frac{\text{d}(\delta y)}{\text{d}x}\text{d}x\\ =& {\int }_{x_0}^{x_1}\frac{\partial F}{\partial y}\delta y\text{d}x+{\int }_{x_0}^{x_1}\frac{\partial F}{\partial y^{\prime }}\text{d}(\delta y)\\ =& {\int }_{x_0}^{x_1}\frac{\partial F}{\partial y}\delta y\text{d}x+{\frac{\partial F}{\partial y^{\prime }}\delta y|}_{x_0}^{x_0}-{\int }_{x_0}^{x_1}\delta y\text{d}\frac{\partial F}{\partial y^{\prime }}\\ =& {\int }_{x_0}^{x_1}\frac{\partial F}{\partial y}\delta y\text{d}x-{\int }_{x_0}^{x_1}\delta y\frac{\text{d}}{\text{d}x}\left(\frac{\partial F}{\partial y^{\prime }}\right)\text{d}x\\ =& {\int }_{x_0}^{x_1}\left[\frac{\partial F}{\partial y}-\frac{\text{d}}{\text{d}x}\left(\frac{\partial F}{\partial y^{\prime }}\right)\right]\delta y\text{d}x\end{array}$$

当 $I$ 取极值时, 对任意微小变化 $\delta y$ 均有 $\delta I=0$, 则要求
$$\frac{\partial F}{\partial y}-\frac{\text{d}}{\text{d}x}\left(\frac{\partial F}{\partial y^{\prime }}\right)=0$$

此即欧拉-拉格朗日方程, 简称E-L方程. 亦可将导数展开, 得到
$$\frac{\partial F}{\partial y}-\frac{{\partial }^{2}F}{\partial y^{\prime }\partial x}-y^{\prime }\frac{{\partial }^{2}F}{\partial y^{\prime 2}}-y^{\prime \prime }\frac{{\partial }^{2}F}{\partial y^{\prime }\partial y^{\prime \prime }}=0$$

一般地, 高阶函数的E-L方程为
$$\frac{\partial F}{\partial y}-\frac{\text{d}}{\text{d}x}\left(\frac{\partial F}{\partial y^{\prime }}\right)+\frac{{\text{d}}^2}{\text{d}{x}^2}\left(\frac{\partial F}{\partial y^{\prime \prime }}\right)-\cdots =\sum _{i=0}^n(-1{)}^i\frac{{\text{d}}^i}{\text{d}{x}^i}\left(\frac{\partial F}{\partial y^{(i)}}\right)=0$$

在前述的推导中的分部积分中, 可以看到高阶E-L方程中正负交替的显然性.

由此定义变分导数为 $$\frac{\delta F}{\delta y}=\frac{\partial F}{\partial y}-\frac{\text{d}}{\text{d}x}\left(\frac{\partial F}{\partial y^{\prime }}\right)+\frac{{\text{d}}^2}{\text{d}{x}^2}\left(\frac{\partial F}{\partial y^{\prime \prime }}\right)-\cdots =\sum _{i=0}^n(-1{)}^i\frac{{\text{d}}^i}{\text{d}{x}^i}\left(\frac{\partial F}{\partial y^{(i)}}\right)$$

E-L方程重写为 $$\frac{\delta F}{\delta y}=0$$

进一步地,

应用

两点间最短路径

过 $(x_0,y_0)$ 和 $(x_1,y_1)$ 作曲线 $y=f(x)$, 则曲线长
$$I(y)={\int }_{x_0}^{x_1}\sqrt{\text{d}{x}^2+\text{d}{y}^2}={\int }_{x_0}^{x_1}\sqrt{1+y^{\prime 2}}\text{d}x$$

令 $F(x,y,y^{\prime })=\sqrt{1+y^{\prime 2}}$, 则 $\frac{\partial F}{\partial y}=0$, $\frac{\partial F}{\partial y^{\prime }}=\frac{y^{\prime }}{\sqrt{1+y^{\prime 2}}}$, 代入E-L方程, 得到
$$\begin{array}{rl}\frac{\text{d}}{\text{d}x}\frac{y^{\prime }}{\sqrt{1+y^{\prime 2}}}& =0\\ \frac{y^{\prime }}{\sqrt{1+y^{\prime 2}}}& =C\\ y^{\prime }& =C\sqrt{1+y^{\prime 2}}\\ y^{\prime }& =\pm \sqrt{\frac{C^2}{1-C^2}}=k\\ \frac{\text{d}y}{\text{d}x}& =k\\ y& =kx+b\end{array}$$

其中 $k$, $b$ 均为任意常数. $(x_0,y_0)$ 和 $(x_1,y_1)$ 代入方程得到
$$\frac{x_1-x}{x_1-x_0}=\frac{y_1-y}{y_1-y_0}$$

最速降线

过点 $A=(x_0,y_0)$ 和点 $B=(x_1,y_1)$ 作光滑的曲线 $y=f(x)$, 其中 $y_0>y_1$, 则质量为 $m$ 的质点从 $A$ 静止出发做机械运动到 $B$ 的用时的积分推导如下
质点在 $(x,y)$ 处时下落 $y_0-y>0$, 重力势能转为动能, 速度 $v$ 满足
$$\begin{array}{rl}\frac{1}{2}m{v}^2& =mg(y_0-y)\\ v& =\sqrt{2g(y_0-y)}\end{array}$$

同时速度等于路程对时间的导数, 满足 $$v=\frac{\text{d}s}{\text{d}t}=\sqrt{1+y^{\prime 2}}\frac{\text{d}x}{\text{d}t}$$

联立上述两式, 消去 $v$, 得到
$$\begin{array}{rl}\sqrt{2g(y_0-y)}& =\sqrt{1+y^{\prime 2}}\frac{\text{d}x}{\text{d}t}\\ \text{d}t& =\sqrt{\frac{1+y^{\prime 2}}{2g(y_0-y)}}\text{d}x\end{array}$$

用时积分如下
$$T(y)={\int }_{x_0}^{x_1}\text{d}t={\int }_{x_0}^{x_1}\sqrt{\frac{1+y^{\prime 2}}{2g(y_0-y)}}\text{d}x=\frac{1}{\sqrt{2g}}{\int }_{x_0}^{x_1}\sqrt{\frac{1+y^{\prime 2}}{y_0-y}}\text{d}x$$

令 $F(x,y,y^{\prime })=\sqrt{\frac{1+y^{\prime 2}}{y_0-y}}$, 注意到与 $x$ 无关的泛函 $G(y,y^{\prime })$ 有
$$\frac{\text{d}}{\text{d}x}\left(G-y^{\prime }\frac{\partial G}{\partial y^{\prime }}\right)=y^{\prime }\frac{\partial G}{\partial y}+y^{\prime \prime }\frac{\partial G}{\partial y^{\prime }}-y^{\prime \prime }\frac{\partial G}{\partial y^{\prime }}-y^{\prime }\frac{\text{d}}{\text{d}x}\frac{\partial G}{\partial y^{\prime }}=y^{\prime }\frac{\delta G}{\delta y}$$

则E-L方程化为
$$\frac{\delta F}{\delta y}=0=y^{\prime }\frac{\delta F}{\delta y}=\frac{\text{d}}{\text{d}x}\left(F-y^{\prime }\frac{\partial F}{\partial y^{\prime }}\right)$$

$$F-y^{\prime }\frac{\partial F}{\partial y^{\prime }}=\sqrt{\frac{1+y^{\prime 2}}{y_0-y}}-\frac{y^{\prime 2}}{\sqrt{(y_0-y)(1+y^{\prime 2})}}=C_0$$

$$1+y^{\prime 2}-y^{\prime 2}=C_0\sqrt{(y_0-y)(1+y^{\prime 2})}$$

$$C_1=(y_0-y)(1+y^{\prime 2})$$

$$\frac{\text{d}y}{\text{d}x}=-\sqrt{\frac{C_1}{y_0-y}-1}$$

$$\text{d}x=\sqrt{\frac{y_0-y}{C_1-(y_0-y)}}\text{d}(y_0-y)$$

做换元 $y_0-y=C_1{\sin }^2\frac{\theta }{2}=\frac{1}{2}{C}_1(1-\cos \theta )$
$$\begin{array}{rl}\text{d}x& =\sqrt{\frac{C_1{\sin }^2\frac{\theta }{2}}{C_1-C_1{\sin }^2\frac{\theta }{2}}}\text{d}{C}_1{\sin }^2\frac{\theta }{2}\\ \text{d}x& =\frac{\sin \frac{\theta }{2}}{\cos \frac{\theta }{2}}\cdot 2C_1\sin \frac{\theta }{2}\cos \frac{\theta }{2}\text{d}\frac{\theta }{2}\\ \text{d}x& =2C_1{\sin }^2\frac{\theta }{2}\text{d}\frac{\theta }{2}\\ \text{d}x& =\frac{1}{2}{C}_1(1-\cos \theta )\text{d}\theta \\ x& =\frac{1}{2}{C}_1(\theta -\sin \theta )+C_2\end{array}$$

代入初值, 解得 $$\{\begin{array}{l}x=x_0+k(\theta -\sin \theta )\\ y=y_0-k(1-\cos \theta )\end{array}$$ 其中 $k$ 使方程组有解
$$\{\begin{array}{l}{x}_1=x_0+k({\theta }_1-\sin {\theta }_1)\\ y_1=y_0-k(1-\cos {\theta }_1)\end{array}$$

比较解与摆线定义, 容易发现最速降线是摆线.
另外, 最速降线过 $B$ 点的切线未必是水平线, 对解求微分得到
$$\{\begin{array}{l}\text{d}x=k(1-\cos \theta )\text{d}\theta \\ \text{d}y=-k\sin \theta \text{d}\theta \end{array}$$

当过 $B$ 点的切线为水平线时, 有
$$\begin{gathered} {\text{d}y|}_{(x,y)=(x_1,y_1)}=0 \\ -k\sin {\theta }_1\text{d}\theta =0 \\ {\theta }_1=\pi \end{gathered}$$

代入到 $k$ 的方程组得到
$$\{\begin{array}{l}{x}_1=x_0+\pi k\\ y_1=y_0-2k\end{array}$$

如此, 当过 $B$ 点的最速降线的切线为水平线时, $B$ 点恒在以 $A$ 为起点, 斜率为 $-\frac{2}{\pi }$ 的直线上.