YYnotesYYNotes

学习目标Learning Objectives - Note

#学习目标#关键概念#定义与公式#推导

第2周:割线法、Muller法与差商结构Week 2 - Secant, Muller, and Divided Differences

本笔记将第2周关于无导数迭代与多项式重构的内容进行规范化整理。Bilingual notes on derivative-free root finding and interpolation structure.

本笔记将第2周关于无导数迭代(Iteration)与多项式重构的内容进行规范化整理。

These notes formalize Week 2 topics on derivative-free iteration and polynomial reconstruction.

学习目标

Learning Objectives

  • 理解牛顿法(Newton's Method)误差(Error)行为,并解释其局部二次收敛(Convergence)原因。

  • Interpret Newton error behavior and explain why local convergence is quadratic.

  • 在导数不可得或计算代价高时使用割线法(Secant Method)

  • Use the secant method when derivatives are unavailable or expensive.

  • Construct second-order models for Muller-type updates.

  • 构造二阶模型并用于 Muller 型更新。

  • 使用牛顿差商(Divided Difference)构造插值多项式。

  • Use Newton divided differences to build interpolation polynomials.

关键概念

Key Concepts

  • 统一误差(Error)记号:en=xnαe_n=x_n-\alpha

  • Standard error notation: en=xnαe_n=x_n-\alpha.

  • 割线法(Secant Method)用两点斜率近似导数。

  • Secant method approximates derivative by slope between two points.

  • 差商(Divided Difference)记号:f[xi,xj]=f(xj)f(xi)xjxif[x_i,x_j]=\dfrac{f(x_j)-f(x_i)}{x_j-x_i}

  • Divided difference notation: f[xi,xj]=f(xj)f(xi)xjxif[x_i,x_j]=\dfrac{f(x_j)-f(x_i)}{x_j-x_i}.

  • 二阶差商(Divided Difference)给出牛顿形式二次项的主系数。

  • Second divided difference gives quadratic leading coefficient in Newton form.

定义与公式

Definitions and Formulas

简单根附近的牛顿误差(Error)模型:en+1f(α)2f(α)en2e_{n+1}\approx \dfrac{f''(\alpha)}{2f'(\alpha)}e_n^2

Newton error model near a simple root: en+1f(α)2f(α)en2e_{n+1}\approx \dfrac{f''(\alpha)}{2f'(\alpha)}e_n^2.

割线迭代(Iteration)公式:

Secant iteration:

xn+1=xnf(xn)xnxn1f(xn)f(xn1).x_{n+1}=x_n-f(x_n)\dfrac{x_n-x_{n-1}}{f(x_n)-f(x_{n-1})}.

Quadratic interpolation (Muller model): p2(x)=c0+c1x+c2x2p_2(x)=c_0+c_1x+c_2x^2 fitted to three samples.

二次插值(Muller 模型):用三点拟合 p2(x)=c0+c1x+c2x2p_2(x)=c_0+c_1x+c_2x^2

牛顿插值(二次形式):

Newton interpolation (quadratic form):

p2(x)=f[x0]+f[x0,x1](xx0)+f[x0,x1,x2](xx0)(xx1).p_2(x)=f[x_0]+f[x_0,x_1](x-x_0)+f[x_0,x_1,x_2](x-x_0)(x-x_1).

推导

Derivations

牛顿误差(Error)估计(课堂回顾)

Newton Error Estimate (Lecture Review)

xn+1=xnf(xn)f(xn)x_{n+1}=x_n-\dfrac{f(x_n)}{f'(x_n)} 出发,在 xnx_n 附近对 f(α)=0f(\alpha)=0 作泰勒展开。

Starting from xn+1=xnf(xn)f(xn)x_{n+1}=x_n-\dfrac{f(x_n)}{f'(x_n)}, expand f(α)=0f(\alpha)=0 by Taylor around xnx_n.

一阶项相消后,主导项与 en2e_n^2 成正比。

After cancellation of first-order terms, the dominant term is proportional to en2e_n^2.

因此牛顿法(Newton's Method)在简单根附近具有二阶精度。

Hence Newton has second-order precision near a simple root.

由两点连线推导割线更新

Secant Update from Line Through Two Points

用两点 (xn1,fn1)(x_{n-1},f_{n-1})(xn,fn)(x_n,f_n) 定义直线。

Use points (xn1,fn1)(x_{n-1},f_{n-1}) and (xn,fn)(x_n,f_n) to define a line.

令直线值为零,得到下一次根估计。

Set the line value to zero to approximate the next root estimate.

即可得到上面的割线公式。

This yields the secant formula above.

差商(Divided Difference)得到主系数

Leading Coefficient via Divided Differences

For f(x)=a+bx+cx2f(x)=a+bx+cx^2, we have f[x0,x1,x2]=cf[x_0,x_1,x_2]=c.

f(x)=a+bx+cx2f(x)=a+bx+cx^2,有 f[x0,x1,x2]=cf[x_0,x_1,x_2]=c

这说明二阶差商(Divided Difference)刻画了二次曲率信息。

This explains why second divided difference encodes quadratic curvature.

例题精讲

Worked Examples

例题模块1:2\sqrt{2} 的牛顿误差(Error)

Example Block 1: Newton Error for 2\sqrt{2}

Take f(x)=x22f(x)=x^2-2 with x0=1.4x_0=1.4.

f(x)=x22f(x)=x^2-2,初值 x0=1.4x_0=1.4

Then x1=1.41.4222(1.4)=1.4+1701.4142857x_1=1.4-\dfrac{1.4^2-2}{2(1.4)}=1.4+\dfrac{1}{70}\approx1.4142857.

x1=1.41.4222(1.4)=1.4+1701.4142857x_1=1.4-\dfrac{1.4^2-2}{2(1.4)}=1.4+\dfrac{1}{70}\approx1.4142857

修正量与二次误差(Error)下降规律一致。

The correction magnitude is consistent with a quadratic error drop.

例题模块2:x22=0x^2-2=0 的割线一步

Example Block 2: Secant Step for x22=0x^2-2=0

Use x0=1.8x_0=1.8 and x1=1.6x_1=1.6.

x0=1.8x_0=1.8x1=1.6x_1=1.6

Compute x2=x1f(x1)x1x0f(x1)f(x0)1.4137931x_2=x_1-f(x_1)\dfrac{x_1-x_0}{f(x_1)-f(x_0)}\approx1.4137931.

计算得 x2=x1f(x1)x1x0f(x1)f(x0)1.4137931x_2=x_1-f(x_1)\dfrac{x_1-x_0}{f(x_1)-f(x_0)}\approx1.4137931

例题模块3:由数据恢复二次多项式

Example Block 3: Recover Quadratic from Data

给定 (3,7)(3,7)(4,6)(4,6)(5,9)(5,9),求 f(x)=a+bx+cx2f(x)=a+bx+cx^2

Given (3,7)(3,7), (4,6)(4,6), (5,9)(5,9), solve for f(x)=a+bx+cx2f(x)=a+bx+cx^2.

差商(Divided Difference)f[3,4]=1f[3,4]=-1f[4,5]=3f[4,5]=3f[3,4,5]=2f[3,4,5]=2

Using divided differences, f[3,4]=1f[3,4]=-1, f[4,5]=3f[4,5]=3, f[3,4,5]=2f[3,4,5]=2.

牛顿形式为 f(x)=7(x3)+2(x3)(x4)f(x)=7-(x-3)+2(x-3)(x-4)

Newton form is f(x)=7(x3)+2(x3)(x4)f(x)=7-(x-3)+2(x-3)(x-4).

误差(Error)分析

Error Analysis

  • 二分法(Bisection Method):一阶线性收敛(Convergence)且区间包根有保证。

  • Bisection: first-order linear convergence with guaranteed bracketing.

  • 牛顿法(Newton's Method):局部二阶收敛(Convergence),但依赖导数。

  • Newton: second-order local convergence but derivative-dependent.

  • 割线法(Secant Method):在光滑情形下常为超线性收敛(Convergence),通常慢于牛顿法(Newton's Method)但不需导数。

  • Secant: superlinear convergence in many smooth cases, usually slower than Newton but derivative-free.

  • 二次插值类方法精度高,但对采样点分布与舍入误差(Error)更敏感。

  • Quadratic interpolation methods can be accurate but are more sensitive to data geometry and rounding.

常见错误

Common Mistakes

警示模块

Warning Block

  • 在割线公式中把分母顺序写反。

  • Switching denominator order in the secant formula.

  • 差商(Divided Difference)与普通差分混淆。

  • Confusing divided differences with ordinary finite differences.

  • 采样点过近导致灾难性消去误差(Error)

  • Using nearly identical sample points, causing catastrophic cancellation.

  • Assuming every three-point quadratic model gives a stable Muller update.

  • 误以为任意三点二次模型都能稳定地产生 Muller 更新。

总结

Summary

总结模块

Summary Block

  • 第2周将求根从依赖导数的牛顿法(Newton's Method)扩展到无导数的割线法(Secant Method)和二次模型方法。

  • Week 2 extends root-finding from derivative-based Newton to derivative-free secant and quadratic-model ideas.

  • 差商(Divided Difference)框架统一了插值构造与系数解释。

  • The divided-difference framework unifies interpolation construction and coefficient interpretation.

  • 这些工具将求根与插值统一为“局部模型”思想的两种表现。

  • These tools connect root-finding and interpolation as two views of local modeling.

练习题

Practice Questions

  • 从两点直线方程直接推导割线公式。

  • Derive the secant formula directly from the two-point line equation.

  • f(x)=x25f(x)=x^2-5,以 x0=2x_0=2x1=3x_1=3 进行三步割线迭代(Iteration)

  • For f(x)=x25f(x)=x^2-5, perform three secant iterations from x0=2x_0=2, x1=3x_1=3.

  • 用牛顿差商(Divided Difference)重构经过 (2,1),(3,6),(4,5),(5,7)(2,1),(3,6),(4,5),(5,7) 的三次多项式。

  • Using Newton divided differences, reconstruct the cubic through (2,1),(3,6),(4,5),(5,7)(2,1),(3,6),(4,5),(5,7).