Piecewise linear secant approximation via Algorithmic Piecewise Differentiation. Griewank, A., Streubel, T., Lehmann, L., Radons, M., & Hasenfelder, R. August, 2017. 10 citations (Semantic Scholar/arXiv) [2023-07-14] arXiv:1701.04368 [math]
Piecewise linear secant approximation via Algorithmic Piecewise Differentiation [link]Paper  abstract   bibtex   
It is shown how piecewise differentiable functions $F: {\}mathbb R{\textasciicircum}n {\}mapsto {\}mathbb R{\textasciicircum}m $ that are defined by evaluation programs can be approximated locally by a piecewise linear model based on a pair of sample points ${\}check x$ and ${\}hat x$. We show that the discrepancy between function and model at any point $x$ is of the bilinear order $O({\}{\textbar}x-{\}check x{\}{\textbar} {\}{\textbar}x-{\}hat x{\}{\textbar})$. This is a little surprising since $x {\}in {\}mathbb R{\textasciicircum}n$ may vary over the whole Euclidean space, and we utilize only two function samples ${\}check F=F({\}check x)$ and ${\}hat F=F({\}hat x)$, as well as the intermediates computed during their evaluation. As an application of the piecewise linearization procedure we devise a generalized Newton's method based on successive piecewise linearization and prove for it sufficient conditions for convergence and convergence rates equaling those of semismooth Newton. We conclude with the derivation of formulas for the numerically stable implementation of the aforedeveloped piecewise linearization methods.
@misc{griewank_piecewise_2017,
	title = {Piecewise linear secant approximation via {Algorithmic} {Piecewise} {Differentiation}},
	url = {http://arxiv.org/abs/1701.04368},
	abstract = {It is shown how piecewise differentiable functions \$F: {\textbackslash}mathbb R{\textasciicircum}n {\textbackslash}mapsto {\textbackslash}mathbb R{\textasciicircum}m \$ that are defined by evaluation programs can be approximated locally by a piecewise linear model based on a pair of sample points \${\textbackslash}check x\$ and \${\textbackslash}hat x\$. We show that the discrepancy between function and model at any point \$x\$ is of the bilinear order \$O({\textbackslash}{\textbar}x-{\textbackslash}check x{\textbackslash}{\textbar} {\textbackslash}{\textbar}x-{\textbackslash}hat x{\textbackslash}{\textbar})\$. This is a little surprising since \$x {\textbackslash}in {\textbackslash}mathbb R{\textasciicircum}n\$ may vary over the whole Euclidean space, and we utilize only two function samples \${\textbackslash}check F=F({\textbackslash}check x)\$ and \${\textbackslash}hat F=F({\textbackslash}hat x)\$, as well as the intermediates computed during their evaluation. As an application of the piecewise linearization procedure we devise a generalized Newton's method based on successive piecewise linearization and prove for it sufficient conditions for convergence and convergence rates equaling those of semismooth Newton. We conclude with the derivation of formulas for the numerically stable implementation of the aforedeveloped piecewise linearization methods.},
	urldate = {2023-07-12},
	publisher = {arXiv},
	author = {Griewank, Andreas and Streubel, Tom and Lehmann, Lutz and Radons, Manuel and Hasenfelder, Richard},
	month = aug,
	year = {2017},
	note = {10 citations (Semantic Scholar/arXiv) [2023-07-14]
arXiv:1701.04368 [math]},
	keywords = {/unread, Mathematics - Numerical Analysis},
}

Downloads: 0