Extension of the Total Least Square Problem Using General Unitarily Invariant Norms
Chi-Kwong Li,Xin-Guo Liu,Xue-Feng Wang
DOI: https://doi.org/10.1080/03081080600593428
2007-01-01
Abstract:Let m, n,p be positive integers such that m >= n + p. Suppose (A, B) is an element of C-mxn x C-mxp, and letP(A,B) = {(EF) is an element of C-mxn x C-mxp: there is X is an element of C-nxp such that (A -E)X=B-F}.The total least square problem concerns the determination of the existence of (E, F) in P(A, B) having the smallest Frobenius norm. In this article, we characterize elements of the set P(A, B) and derive a formula forrho(A,B) = inf{parallel to[E\F]parallel to: (E,F) is an element of P(A,B)},for any unitarily invariant norm parallel to.parallel to on Cmx(n+p), where [E\F] denotes the m x (n + p) matrix formed by the columns of E and F. Furthermore, we give a necessary and sufficient condition on (A, B) and the unitarily invariant norm parallel to.parallel to so that there exists (E, F) is an element of P(A, B) attaining rho(A,B). The results cover those on the total least square problem, and those of Huang and Yan on the existence of (E, F) is an element of P(A, B) so that [E\F] has the smallest spectral norm.