On a greedy algorithm in the space \(L_p[0,1]\) (Q1033990)

Let \(X\) be a Banach space. A subset \(\mathcal D\) of \(S_X\) (the unit sphere of \(X\)) is called a dictionary if the closed linear space generated by \(\mathcal D\) is \(X\). For \(f,g\in X\) with \(\|g\|=1\), let \( \mu(f,g)\in \mathbb R\) be such that \(\|f-\mu(f,g)\,g\|=\inf_\mu \|f-\mu\,g\| \) and let \(r(f,g)=\|f\|- \|f-\mu(f,g)\,g\|\). The author considers the following greedy algorithm (XGA): for given \(f_0\in X\), define inductively \(g_{m+1}\in \mathcal D\) such that (1)\; \(r(f_m,g_{m+1})=\sup_{g\in \mathcal D}r(f_m,g)\) and set \(f_{m+1}=f_m-\mu(f_m,g_{m+1})\,g_{m+1}. \) The author studies the \(X\)-greedy algorithm in the space \(L_p[0,1]\) when \(\mathcal D\) is formed of functions proportional to the characteristic functions of the dyadic intervals \(\Delta^i_k=[2^{-k}\,(i-1),2^{-k}\,i]\). He gives estimates of \(\|f_m\|\) and proves the convergence of the algorithm in the case \(1<p<\infty\). In Hilbert spaces, the so-called Pure Greedy Algorithm (PGA) always converges, which is not the case in Banach setting. A study of various conditions of convergence of greedy algorithms in Banach spaces is given in [\textit{S.\,J.\thinspace Dilworth, D.\,Kutzarova, K.\,L.\thinspace Shuman, V.\,N.\thinspace Temlyakov} and \textit{P.\,Wojtaszczyk}, J.~Fourier Anal.\ Appl.\ 14, No.\,5--6, 609--628 (2008; Zbl 1180.41030)], see also the author's paper [Math.\ Notes 73, No.\,3, 342--358 (2003); translation from Mat.\ Zametki 73, No.\,3, 371--389 (2003; Zbl 1058.41036)] and the survey by \textit{V.\,N.\thinspace Temlyakov} [Acta Numerica 17, 235--409 (2008; Zbl 1178.65050)].