学习目标Learning Objectives - Note
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:。
-
Standard error notation: .
-
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:。
-
Divided difference notation: .
-
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:。
Newton error model near a simple root: .
割线迭代(Iteration)公式:
Secant iteration:
Quadratic interpolation (Muller model): fitted to three samples.
二次插值(Muller 模型):用三点拟合 。
牛顿插值(二次形式):
Newton interpolation (quadratic form):
推导
Derivations
牛顿误差(Error)估计(课堂回顾)
Newton Error Estimate (Lecture Review)
由 出发,在 附近对 作泰勒展开。
Starting from , expand by Taylor around .
一阶项相消后,主导项与 成正比。
After cancellation of first-order terms, the dominant term is proportional to .
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 与 定义直线。
Use points and 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.
用差商(Divided Difference)得到主系数
Leading Coefficient via Divided Differences
For , we have .
对 ,有 。
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: 的牛顿误差(Error)
Example Block 1: Newton Error for
Take with .
取 ,初值 。
Then .
则 。
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: 的割线一步
Example Block 2: Secant Step for
Use and .
取 、。
Compute .
计算得 。
例题模块3:由数据恢复二次多项式
Example Block 3: Recover Quadratic from Data
给定 、、,求 。
Given , , , solve for .
用差商(Divided Difference)得 、、。
Using divided differences, , , .
牛顿形式为 。
Newton form is .
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.
-
将差商(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.
-
对 ,以 、 进行三步割线迭代(Iteration)。
-
For , perform three secant iterations from , .
-
用牛顿差商(Divided Difference)重构经过 的三次多项式。
-
Using Newton divided differences, reconstruct the cubic through .