人工智能教程 - 数学基础课程1.2 - 数学分析(三)-1-2

ODE几何方法

ODE:ordinary differential equation 常微分方程

First-order ODE’s

y’ = f(x, y)

for example:
$y^{\prime }=\frac{x}{y};y^{\prime }=y-x^2$
$y^{\prime }=x-y^2$

GEOM:View of ODE’s (几何上讲)
Analytic:
y’ = f(x, y) $\iff$ Geometric direction field
$y_1(x)$ solution $\iff$ Integral curve: slope = f(x, y)

其每点处的方向与方向场相同(has the direction of the field everywhere at all points)

$y_1(x)$ solution to y’ = f(x, y) $\iff$ graph of $y_1(x)$ is an integral curve
int curves:cirles

用分离变量法求解ODE

1. Two integral curve can’t cross
2. Two integral curve can’t be tangent touch

UNIQ: has one and only one solution

欧拉数值方法及推广

(Numerical solutions)

IVP: $\{\begin{array}{c}{y}^{\prime }=f(x,y)\\ y_0(x_0)=y_0\end{array}$

欧拉方法(Euler’s Method)

h = step size

$y_{n+1}-y_n=h{A}_n$
$y_{n+1}=y_n+h{A}_n$

Euler’s equations

$x_{n+1}=x_n+h$
$y_{n+1}=y_n+h{A}_n$
$A_n=f(x_n,y_n)$

Ex:
$y^{\prime }=x^2-y^2$ ; y(0) = 1 ; h = .1

n	$x_n$	$y_n$	$A_n$	$h{A}_n$
0	0	1	-1	-.1
1	.1	.9	-.80	-.08
2	.2	.82

convex y’’>0 E: too low

concave y’’<0 E: too high

Better method

• smaller step size
• find a better slope
  better value than $A_n$
  $X_{n+1}=x_n+b$
  $y_{n+1}=y_n+h(\frac{A_n+B_n}{2})$
  $\widetilde{y_{n+1}}=y_n+h{A}_n$
  Henn’s - improved Euler

1, PK2:二阶方法 second - order

$e\approx c_2{h}^2$

2, PK4:standard method

-accurate 准确的龙格-库塔(Runge-kutta)四阶方法

3, 加权平均 super slope

$\frac{A_n+2B_n+2C_n+D_n}{6}$