Generalized cross approximation for 3D-tensors (Q434216)

The algorithm CA2D described by \textit{J. Schneider} [J. Approx. Theory 162, No. 9, 1685--1700 (2010; Zbl 1208.41017)] for two-dimensional cross approximation is generalized to the three-dimensional case and reformulated for 3D-tenors. In terms of a trivariate function \(f(x,y,z)\) defined on \([0,1]^3\), it computes in an outer loop the recurrence \[ R_k(x,y,z)=R_{k-1}(x,y,z)- {R_{k-1}(x,y_k,z)R_{k-1}(x_k,y,z) \over R_{k-1}(x_k,y_k,z)}, ~~\mathrm{for}~~k=1,2,\ldots, \] starting from \(R_0=f\). The pivot \((x_k,y_k)\) is such that the denominator is nonzero for all \(z\). After \(T\) iterations \(f-R_T\) is written as a sum of \(T\) of the latter quotients. The bivariate functions in the numerator are approximated by CA2D in an inner loop. Error analysis, a pivoting strategy, and the computational complexity of this method are analysed.