On the failure of proximinality of tensor-product subspaces (Q2640818): Difference between revisions

Let \(S=T=[-1,1]\). The authors construct a function \(f\in C(S\times T)\), endowed with the uniform norm, which fails to have an element of best approximation in the subspace \(W=\{w(x,t)=x_ 0(s)+tx_ 1(s)+y_ 0(t)+sy_ 1(t)\}.\) Note that \(W=C(S)\otimes H+G\otimes C(T),\) where G and H are two-dimensional spaces of polynomials of degree 1. Earlier work on this problem by \textit{S. P. Diliberto} and \textit{E. G. Straus} [Pac. J. Math. 1, 195-210 (1951; Zbl 0043.068)] had shown that the subspace \(\{x(s)+y(t):\) x,y\(\in C[-1,1]\}\) is proximinal in C(S\(\times T)\); this was later extended to subspaces of the form \(\{x(s)+\sum^{n}_{j=1}s^{j-1}y_ j(t):\) \(x,y_ 1,...,y_ n\in C[-1,1]\}\) by \textit{E. W. Cheney} and \textit{J. R. Respess} jun. [Pac. J. Math. 102, 437-466 (1982; Zbl 0536.41035)]. Moreover, the second author showed in Approximation theory IV, Proc. int. Conf. Tex. A \& M Univ. 1983, 475-481 (1983; Zbl 0624.41022) that any \(f\in C(S\times T)\) has a best approximation in the above subspace W provided \(\partial f/\partial s\) exists on \(\{\) \(1\}\times T\) and \(\partial f/\partial s(1,\cdot)\in C(T)\).