Error estimates for two-dimensional cross approximation (Q606674)

The author studies the approximation of a function \(f:[0,1]^2\rightarrow\mathbb R\) by special bilinear forms, i.e., by a finite sum of tensor products of one-dimensional functions: \[ \sum^k_{i=1}(g_i\otimes h_i)(x,y)=\sum^k_{i=1}g_i(x)h_i(t). \] The functions \(g_i\) and \(h_i\) are the restrictions of \(f\) itself to certain segments in \([0,1]^2\) and are constructed by an algorithm (named \(CA2D\)), described in section \(1.1\). After \(k\) steps, the algorithm produces a finite sum \(F_k(x,y)\) such that \(f|_{G_k}(x,y)=F_k|_{G_k}(x,y)\), where \(G_k\) is a set of segments in parallel with \(Ox\) and respectively \(Oy\). The main goal is to estimate the error \(R_{k}(x,y):=f(x,y)-F_k(x,y)\), for all \((x,y)\in [0,1]^2\). He discusses the influence of smoothness and structural properties of \(f\) over \(R_k(x,y)\). Also, the choices of the parallel segments in \([0,1]^2\) are studied to obtain estimates for \(|R_k(x,y)|\) in terms of the best approximation of \(f(x,y)\) by arbitrary bilinear forms. Some connections with other results in approximation theory are established too.