数学物理方法 11 变分法

$\color{blue}{第十一章 变分法}$

$\color{blue}{\S 11.1 泛函和泛函的极值}$

$\color{blue}{11.1.1 泛函}$

1696年,Basel大学Bernoulli提出,最速落径:
$v = \dfrac{ds}{dt} = \sqrt{2gy} \quad \therefore 总下降时间$
$T = \int_{t_1(A)}^{t_2(B)}dt = \int_A^B \dfrac{ds}{v},由于ds = \sqrt{dx^2 + dy^2} = \sqrt{1 + (y^{\prime})^2}dx$
$\therefore T = \int_A^B \dfrac{\sqrt{1 +(y^{\prime})^2} dx}{\sqrt{2gy}}$
$即T = T[y(x)] = \int_A^B \dfrac{\sqrt{1 + (y^{\prime})^2} dx}{\sqrt{2gy}}, -是函数的函数$

$1.定义$
$泛函是函数的函数。即J = J[y(x)]。其中,J \in B:复(实);数集y(x) \in C:复数集.$
$注意:(1)不同于普通函数$
$(2)不同于复合函数$
$(3)定义域:y(x)可取类$

$2.典型表达方式:J[y(x)] = \int_a^b F(x, y, y^{\prime}) dx \to (1)$
$F(x, y, y^{\prime})泛函的核$

$3.最速落径问题:求泛函T[y(x)]的极小值问题。$

$\color{blue}{11.1.2 泛函的极值}$

类似于上述的求极值问题，如光学中费马原理，力学中最小作用力问题。在物理学中很多，我们将会看到求泛函极值可归结为两种方法.

$1.求泛函极值方法$
$(1)直接方法:从泛函直接求极值曲线$
$(2)间接法:化为解微分方程-与变分问题有联系$
$为此，建立有关变分概念。$

$2.变分$
$(1):函数的变分:若y(x) \stackrel{微变}{-\to}y(x) + t \eta(x), t为小参数,\\ 则记\delta(y) = t \eta(x) \quad (2) – 称t \eta(x) 为y(x)的变分.\\ 注意:\delta y 不同于dy, dy有一取极值过程, \delta y不取极限,略去了高阶小量.此时，$
$y^{\prime}(x) = \lim_{\Delta x \to 0} \dfrac{\Delta y}{\Delta x} \to \lim _{\Delta x \to 0} \dfrac{\Delta(y + t \eta)}{\Delta x} = y^{\prime}(x) + t \eta^{\prime}(x)$
$(2)变分与微分可交换次序:$
$\delta(y^{\prime}) = t \eta^{\prime}(x) = \dfrac{d}{dx}[t \eta(x)] = \dfrac{d}{dx}\delta(y) \to y^{\prime}(x)的变分$
$即:\delta(y^{\prime}) \equiv \dfrac{d}{dx}\delta(y)$

$(3)：函数的变分$
$若(1)中F \in C^2,y \in C^2, 则当y \to y + t \eta$
$\Delta J = J[y(x) + t \eta] - J[y]$
$=\int_a^b[F(x, y + t \eta, y^{\prime} + t \eta^{\prime}) - F(x, y, y^{\prime})] dx$
$=\int_a^b [\dfrac{\partial F}{\partial y} t \eta + \dfrac{\partial F}{\partial y^{\prime}} t \eta^{\prime} + t的高阶小量] dx$
$即\delta J = \int_a^b(\dfrac{\partial F}{\partial y} \delta y + \dfrac{\partial F}{\partial y^{\prime}} \delta y^{\prime}) dx \to (4) -称为泛函J[y(x)]的第一次变分$

$\color{blue}{11.1.3 泛函极值的必要条件}$

$设J[y(x)]的极值函数为y(x),且有变分t \eta(x),则J[y(x) + t \eta(x)] = \phi(t)\\ (因为y(x)已经设定),由一元函数取极值条件,则J极值条件$
$\to \phi(t)极值条件\to \dfrac{\phi}{dt} |_{t=0} = 0$
$\dfrac{\partial J[y(x) + t \eta(x)]}{\partial t} | _ {t=0} = 0,即, \int_a^b \dfrac{\partial F(x,y+t \eta, y^{\prime} + t \eta^{\prime}]}{\partial t} | _ {t=0} dx = 0$
$\int_a^b [\dfrac{\partial F}{\partial (y+t \eta)}\eta + \dfrac{\partial F}{\partial(y^{\prime} + t \eta^{\prime})} \eta^{\prime}] | _ {t=0} dx = 0,$
$即:\int_a^b [\dfrac{\partial F}{\partial y} \eta + \dfrac{\partial F}{\partial y^{\prime}} \eta^{\prime}] dx = 0 \to \int_a^b [ \dfrac{\partial F}{\partial y} \delta y + \dfrac{\partial F}{\partial y^{\prime}} \delta y^{\prime}] dx = 0 \quad (5)$

$1.泛函极值的问题是变分问题$
$对照(4):极值条件\delta J = 0有$
$\because \int_a^b \dfrac{\partial F}{\partial y^{\prime}} \delta y^{\prime} dx = \int_a^b \dfrac{\partial F}{\partial y^{\prime}} \dfrac{d}{dx}(\delta y) dx$
$= \dfrac{\partial F}{\partial y^{\prime}} \delta y |_a^b - \int_a^b \dfrac{d}{dx}(\dfrac{\partial F}{\partial y^{\prime}}) \delta y dx = -\int_a^b \dfrac{d}{dx}(\dfrac{\partial F}{\partial y^{\prime}}) \delta y dx$
$代入(5):\int_a^b [\dfrac{\partial F}{\partial y} - \dfrac{d}{dx}(\dfrac{\partial F}{\partial y^{\prime}})] \delta y dx = 0$
$\therefore \dfrac{\partial F}{\partial y} - \dfrac{d}{dx}(\dfrac{\partial F}{\partial y^{\prime}}) = 0 \to Euler方程$

$2.泛函取极值的条件-y(x)满足Euler方程$
$(1)对于J[y(x)] = \int_a^b F(x,y,y^{\prime}) dx$
$有\dfrac{\partial F}{\partial y} - \dfrac{d}{dx}(\dfrac{\partial F}{\partial y^{\prime}}) = 0 \to 二阶常微分方程 \Rightarrow y^{\prime}\dfrac{\partial F}{\partial y^{\prime}} - F = C$

$(2)对于J[y_1(x), y_2(x), \cdots, y_n(x)] = \int_a^b F(x;y_1, y_2, \cdots, y_n; y_1^{\prime}, y_2^{\prime}, \cdots, y_n^{\prime}) dx$
$有\dfrac{\partial F}{\partial y_i} - \dfrac{d}{dx}(\dfrac{\partial F}{\partial y_i^{\prime}}) = 0, (i = 0, 1, 2, \cdots, n)$

$(3)对于J[y^{\prime}(x), y^{\prime}(x) \cdots y^{(n)}(x)] = \int_a^b F(x;y; y^{\prime}, y^{\prime \prime} \cdots y^{(n)}) dx$
$有\dfrac{\partial F}{\partial y} - \dfrac{d}{dx}(\dfrac{\partial F}{\partial y^{\prime}}) + \dfrac{d^2}{d x^2}(\dfrac{\partial F}{\partial y^{\prime \prime}}) - \dfrac{d^3}{d x^3}(\dfrac{\partial F}{\partial y^{\prime \prime \prime}}) = 0$

$(4)多元函数J[u(x, y, z)] = \int_a^b F(x, y, z; u; u_x, u_y, u_z)dx dy dz$
$\dfrac{\partial F}{\partial u} - \dfrac{\partial}{\partial x}(\dfrac{\partial F}{\partial u_x}) - \dfrac{\partial}{\partial y}(\dfrac{\partial F}{\partial u_y}) - \dfrac{\partial}{\partial z}(\dfrac{\partial F}{\partial u_z}) = 0$
$于是求泛函数极值问题 \to 解Euler方程问题$

$3.若F不显含x则$
$\dfrac{\partial F}{\partial y} - \dfrac{d}{dx}(\dfrac{\partial F}{\partial y^{\prime}}) = 0 \to y^{\prime} \dfrac{\partial F}{\partial y^{\prime}} - F = C$
$考虑\dfrac{d}{dx}[y^{\prime} \dfrac{\partial F}{\partial y^{\prime}} - F] = y^{\prime \prime} \dfrac{\partial F}{\partial y^{\prime}} + y^{\prime} \dfrac{d}{dx}(\dfrac{\partial F}{\partial y^{\prime}}) - \dfrac{\partial F}{\partial x} - \dfrac{\partial F}{\partial y}y^{\prime} - \dfrac{\partial F}{\partial y^{\prime}} y^{\prime \prime}$
$= -y^{\prime}[\dfrac{\partial F}{\partial y} - \dfrac{d}{dx}(\dfrac{\partial F}{\partial y^{\prime}})] - \dfrac{\partial F}{\partial x}$
$\because F不显示含x,即\dfrac{\partial F}{\partial x} = 0 \Rightarrow \dfrac{d}{dx}[y^{\prime} \dfrac{\partial F}{\partial y^{\prime}} - F] = 0,$
$\therefore 积分可得y^{\prime} \dfrac{\partial F}{\partial y^{\prime}} - F = C(B)$

$例1.求最速落径$
$T[y(x)] = \int_{x_A}^{x_B} \dfrac{\sqrt{1 + (y^{\prime})^2} dx}{\sqrt{2gy}}$
$\therefore \delta \int_{x_A}^{x_B} \dfrac{\sqrt{1 + (y^{\prime})^2} dx}{\sqrt{2gy}} = 0$
$F = \dfrac{\sqrt{1 + (y^{\prime})^2}dx}{\sqrt{2gy}} = \dfrac{1}{\sqrt{2g}}\dfrac{\sqrt{1 +(y^{\prime})^2}}{\sqrt{y}} 不显含x$
$于是Eulerf方程$
$y^{\prime} \cdot \dfrac{\partial}{\partial y^{\prime}} \dfrac{\sqrt{1 + (y^{\prime})^2}}{\sqrt{y}} - \dfrac{\sqrt{1 + (y^{\prime})^2}}{\sqrt{y}} = C$
$\dfrac{(y^{\prime})^2}{\sqrt{[1 + (y^{\prime})^2]y}} - \dfrac{\sqrt{1 + (y^{\prime})^2}}{\sqrt{y}} = C$
$\Rightarrow \dfrac{(y^{\prime})^4}{[1 + (y^{\prime})^2]y} + \dfrac{1 + (y^{\prime})^2}{y} - 2 \dfrac{y^{\prime 2}}{y} = C^2$
$\dfrac{1}{[1 + (y^{\prime})^2]y} = C^2, 令\dfrac{1}{C^2} = C_1$
$得y^{\prime} = \dfrac{\sqrt{C_1 - y}}{\sqrt{y}}, \to x - C_2 = \int \dfrac{\sqrt{y}}{\sqrt{C_1 - y}} dy$
$令y = C_1\sin^2\dfrac{\theta}{2},则\left \lbrace \begin{array}{l}x = \dfrac{C_1}{2}(\theta - \sin \theta) + C_2 \\ y = \dfrac{C_1}{2}(1 - \cos \theta) \end{array} \right.$
$是由半径为\dfrac{C_1}{2}的圆周上一固定点运动产生的.$
$在途中x轴下方滚动.存在一条且仅一条通过原点及点(x_b, y_b)的摆线.\\ 适当选择C_1和C_2可以给出这条摆线.$

$\color{blue}{11.1.4 泛函的条件极值}$

$\left \lbrace \begin{array}{l}J[y(x)] = \int_a^b F(x, y, y^{\prime}) dx \\ \int_a^b G(x, y, y^{\prime}) dx = l \end{array} \right., y(a) = y_0, y(b) = y_1$
$拉格朗日(Lagrange)乘子法:$
$考虑:\delta\int_a^b [F(x, y, y^{\prime}) + \lambda G(x, y, y^{\prime})] dx = 0$
$则 \to \dfrac{\partial F}{\partial y} + \lambda \dfrac{\partial G}{\partial y} - \dfrac{d}{dx}[(\dfrac{\partial F}{\partial y^{\prime}}) + \lambda \dfrac{d}{dx}(\dfrac{\partial G}{\partial y^{\prime}})] = 0$
$积分常数C_1, C_2和\lambda可由附加条件定出$

$例2.\left \lbrace \begin{array}{l}J[y(x)] = \int_0^1 y^{\prime 2} dx \\ \int_0^1 y^2 dx = 1 \\ y(0) = 0, y(1) = 1 \end{array} \right.$
$考虑 \delta \int_0^1 (y^{\prime 2} + \lambda y^2) dx = 0$
$不显含x,也可推出一阶Euler方程,此处直接用二阶Euler也不困难:\\ 2 \lambda y - \dfrac{d}{dx}(2y^{\prime}) = 0$
$即:y^{\prime \prime} - \lambda y = 0$
$y = C_1 e^{\sqrt{\lambda} x} + C_2 e^{-\sqrt{\lambda}x}$
$由\left \lbrace \begin{array}{l} y(0) = 0 \\ y(1) = 1 \end{array} \right.得:y_n = C_n \sin n \pi x \quad (n = 1, 2, \cdots)$
$再由\int_0^1 y^2 dx = 1得:C_n = \pm \sqrt{2}, \therefore y_n = \pm \sqrt{2} \sin n \pi x$
$J[y(x)] = \int_0^1 [\dfrac{d}{dx} \pm \sqrt{2} \sin n \pi x]^2 dx = n^2 \pi^2$
$极小值为J[y_1(x)] = \pi^2$

$\color{blue}{11.1.5 里兹方法}$

$1.对于J[f(x)],$
$令y(x) = f(\varphi_1(x), \varphi_2(x) \cdots \varphi_n(x); C_1,C_2 \cdots C_n)$
$则J[f(x)] = \varphi(C_1, C_2, \cdots, C_n)$
$于是当\dfrac{\partial \varphi}{\partial C_i} = 0时,J[f(x)]取极值, i = 1, 2, \cdots, n$

$2.注意$
$(1)f是近似解f(x) = \lim \limits _{n \to \infty} f(\varphi_1, \varphi_2, \cdots, \varphi_n; C_1, C_2, \cdots, C_n)$
$(2)适当选\varphi, f$
$注:一般为多项式三角式为它们的线性组合,满足边界条件$

$3.例:求J[f(x)] = \int_0^1 y^{\prime 2} dx(的极小值) (1)$
$其中\left \lbrace \begin{array}{l} \int_0^1 y^2 dx = 1 \\ y(0) = 0, y(1) = 0 \end{array} \right. (2)$
$解:选\varphi_n(x) = C_n x^n$
$令y(x) = x(x-1)(C_0 +C_1x) = C_1x^3 + (C_0 - C_1)x^2 - C_0x \quad (3)$
$代入(1):\phi = J[y(x)] = \int_0^1 [3C_1x^2 + 2(C_0 - C_1)x - C_0]^2 dx$
$= \dfrac{1}{3}(C_0^2 + C_0C_1 + dfrac{2}{5}C_1^2) \quad (4),$
$即\phi = \phi(C_0, C_1)$
$代入(2):\int_0^1 y^2 dx = \dfrac{1}{30}(C_0^2 +C_0C_1 + \dfrac{2}{7}C_1^2) = 1 \quad (5)$
$即\psi = \dfrac{1}{30}(C_0^2 + C_0C_1 + \dfrac{2}{7}C_1^2) - 1 = 0$
$由拉格朗日乘子法:若要求y(x) = f(x_1, x_2, \cdots, x_n)在m个约束条件:$
$g_k(x_1, x_2, \cdots, x_n) = 0, (k = 1, 2, \cdots, m)$
$下的极值只需考虑 F = y + \sum \limits_{k=1}^{m} \lambda_k g_k$
$使\left \lbrace \begin{array}{l}\dfrac{\partial F}{\partial x} = 0 \\ g_k(x_i) = 0 \end{array} \right. i = 1, 2, \cdots, n即可$
$\therefore 应由\left \lbrace \begin{array}{l}\dfrac{\partial F}{\partial C_i} = \dfrac{\partial \phi}{\partial C_i} + \lambda \dfrac{\partial \psi}{\partial C_i} = 0 \\ \varphi = 0 \end{array} \right.联立消去\lambda 来求y$
$但在此不必,由(5):C_0^2 + C_0C_1 = 30 - \dfrac{2}{7}C_1^2 \quad (6)$
$代入(4):\phi = J[y(x)] = \dfrac{1}{3}(30 - \dfrac{2}{7}C_1^2 + \dfrac{2}{5}C_1^2)$
$=\dfrac{2}{3}(15 + \dfrac{2}{35}C_1^2), C_1 = 0有极小值,此时J[f(x)] = 10$
$C_1 = 0代入(6)：C_0 = \pm \sqrt{30},代入(4)$
$\therefore y(x) = \pm \sqrt{30} x(x-1) = \pm \sqrt{30} x(x-1)$
$= \pm \sqrt{30}[(x - \dfrac{1}{2})^2 - \dfrac{1}{4}]$
$(x - \dfrac{1}{2})^2 = \dfrac{1}{\pm \sqrt{30}}y + \dfrac{1}{4} = \pm 2 \dfrac{1}{2 \sqrt{30}}(y \pm \dfrac{\sqrt{30}}{4})$
$令h = \dfrac{1}{2}, k = -(\pm \dfrac{\sqrt{30}}{4}), p = \dfrac{1}{2 \sqrt{30}}$
$\Rightarrow (x-h)^2 = \pm 2 p(y-k)$
$顶点:(h, k); 焦点:(h, k \pm \dfrac{p}{2}); 准线: y = -k \pm \dfrac{p}{2}$

$\color{blue}{\S 11.2 用变分法解数理方程}$

$\color{blue}{11.2.1 变分法}$

$1.思路$
$泛函极值 \leftarrow 泛函的Euler方程的解$
$\qquad \uparrow$
$直接方法求解$

$2.步骤$
$(1)写出定解问题所对应的泛函$
$(2)用直接法(里兹法)求该泛函极值,此即写方程的解.$