Finite computation of the \(\ell_1\) estimator from Huber's \(M\)-estimator in linear regression (Q1885288)

The author reviews previous work on the approximation of linear \(\ell_1\) estimators by Huber's \(M\)-estimator based on the algorithms proposed by \textit{D. I. Clark} and \textit{M. R. Osborne} [SIAM J. Sci. Stat. Comput. 7, 72--85 (1986; Zbl 0593.65100)], and \textit{K. Madsen} and \textit{H. B. Nielsen} [BIT 30, 682--699 (1990; Zbl 0717.65118)]. It is pointed out that although the Madsen-Nielsen algorithm, essentially an extension of the Clark-Osborne algorithm, is a promising one, it is guaranteed to terminate finitely only under certain assumptions. Thus, an extension of the Madsen-Nielson algorithm is given and its finite termination property is proved without any assumptions. Summerized computational experience with the modified algorithm is also provided.