No label defined (Q3976974)

The Iri-Imai unconstrained multiplicative barrier function method [see \textit{M. Iri} and \textit{H. Imai}, ``A method of solving linear programming -- with reference to the Karmarkar method and the penalty function method'', Res. Meet. MP Res. Group OR Soc. Japan, Febr. 16, 1985; and Discrete algorithms and complexity, Proc. Jap.-US Joint Semin., Kyoto/Jap. 1986, Perspect. Comput. 15, 417--435 (1987; Zbl 0648.90050)] to solve the linear programming problem is extended to nonlinear convex problems of class \(C^2: \min\{f(x)\mid g_i(x)\geq 0,\ i=1,\ldots,m\}\), by optimizing sequentially the unconstrained function \(B(x,v) = (s(x)-v)^p(Tg^q_i(x))\), \(p>0\), \(q\in R\), where the operator \(T\) stands for the product \(\left(\prod^m_{i=1}\right)\) or \(\left(\sum^m_{i=1}\right)\) sum operator, \(s(x)=\pm f(x)\) and \(v\) is a parameter updated at each sequence. An algorithm to solve the problem is given, convergence theorems are proved and a numerical example is presented. The above results can also be regarded as a generalization of \textit{P. Huard}'s method of centre [in: Nonlinear Programm., NATO Summer School Menton 1964, 207--220 (1967; Zbl 0157.49701)], by the extension of the range of the parameter \(q\) from \(q>0\) to \(q\in\mathbb{R}\). The authors' further researches [Methods Oper. Res. 64, 31--40 (1991; Zbl 0762.90062)] show that the sequential rate of convergence is decreasing with respect to \(m\), the number of constraints, but, if \(T:=\sum^m_{i=1}\), \(p>0\), \(q<0\), then the sequential convergence rate does not depend on \(m\) and it is better than the convergence rate of a procedure using \(T:=\prod^m_{i=1}\), for any \(m\).