广义 Richardson 外推过程的向量化版本
数值分析
2020-12-08 v3 数值分析
摘要
令 { \xx m } \{\xx_m\} { \xx m } 为满足 \xx m ∼ \sss + ∑ i = 1 ∞ α i ≫ i ( m ) 当 m → ∞ 时 \xx_m\sim \sss+\sum^\infty_{i=1}\alpha_i \gg_i(m)\quad\text{当 $m\to\infty$ 时} \xx m ∼ \sss + i = 1 ∑ ∞ α i ≫ i ( m ) 当 m → ∞ 时 的向量序列,其中 \sss \sss \sss 为 { \xx m } \{\xx_m\} { \xx m } 的极限或反极限(antilimit),且 { ≫ i ( m ) } i = 1 ∞ \{\gg_i(m)\}^\infty_{i=1} { ≫ i ( m ) } i = 1 ∞ 为当 m → ∞ m\to\infty m → ∞ 时的渐近尺度,即 lim m → ∞ ∥ ≫ i + 1 ( m ) ∥ ∥ ≫ i ( m ) ∥ = 0 , i = 1 , 2 , … . \lim_{m\to\infty}\frac{\|\gg_{i+1}(m)\|}{\|\gg_{i}(m)\|}=0,\quad i=1,2,\ldots. m → ∞ lim ∥ ≫ i ( m ) ∥ ∥ ≫ i + 1 ( m ) ∥ = 0 , i = 1 , 2 , … . 向量序列 { ≫ i ( m ) } m = 0 ∞ \{\gg_i(m)\}^\infty_{m=0} { ≫ i ( m ) } m = 0 ∞ (i = 1 , 2 , … i=1,2,\ldots i = 1 , 2 , … )以及 { \xx m } \{\xx_m\} { \xx m } 已知。本文分析了广义 Richardson 外推过程的向量化版本的收敛性与收敛加速性质,该过程由以下方程定义: ∑ i = 1 k ⟨ \yy , Δ ≫ i ( m ) ⟩ α ~ i = ⟨ \yy , Δ \xx m ⟩ , n ≤ m ≤ n + k − 1 ; \sss n , k = \xx n + ∑ i = 1 k α ~ i ≫ i ( n ) , \sum^k_{i=1}\braket{\yy,\Delta\gg_{i}(m)}\widetilde{\alpha}_i=\braket{\yy,\Delta\xx_m},\quad n\leq m\leq n+k-1;\quad \sss_{n,k}=\xx_n+\sum^k_{i=1}\widetilde{\alpha}_i\gg_{i}(n), i = 1 ∑ k ⟨ \yy , Δ ≫ i ( m ) ⟩ α i = ⟨ \yy , Δ \xx m ⟩ , n ≤ m ≤ n + k − 1 ; \sss n , k = \xx n + i = 1 ∑ k α i ≫ i ( n ) , 其中 \sss n , k \sss_{n,k} \sss n , k 为 \sss \sss \sss 的逼近。此处 \yy \yy \yy 为某非零向量,⟨ ⋅ , ⋅ ⟩ \braket{\cdot\,,\cdot} ⟨ ⋅ , ⋅ ⟩ 为内积,满足 ⟨ α \aaa , β \bb ⟩ = α ˉ β ⟨ \aaa , \bb ⟩ \braket{\alpha\aaa,\beta\bb}=\bar{\alpha}\beta\braket{\aaa,\bb} ⟨ α \aaa , β \bb ⟩ = α ˉ β ⟨ \aaa , \bb ⟩ ,且 Δ \xx m = \xx m + 1 − \xx m \Delta\xx_m=\xx_{m+1}-~\xx_m Δ \xx m = \xx m + 1 − \xx m ,Δ ≫ i ( m ) = ≫ i ( m + 1 ) − ≫ i ( m ) \Delta\gg_i(m)=\gg_i(m+1)-\gg_i(m) Δ ≫ i ( m ) = ≫ i ( m + 1 ) − ≫ i ( m ) 。通过对 ≫ i ( m ) \gg_i(m) ≫ i ( m ) 施加最少数量的合理附加条件,我们证明了误差 \sss n , k − \sss \sss_{n,k}-\sss \sss n , k − \sss 当 n → ∞ n\to\infty n → ∞ 时具有完整的渐近展开。我们还证明了实际发生了收敛加速,并给出了其完整分类。
引用
@article{arxiv.1605.02630,
title = {On a Vectorized Version of a Generalized Richardson Extrapolation Process},
author = {Avram Sidi},
journal= {arXiv preprint arXiv:1605.02630},
year = {2020}
}