A combined homotopy interior point method for general nonlinear programming problems (Q1354287)

A system of equations for a Kuhn-Tucker point of a general nonlinear programming problem is treated by means of a numerical homotopy method. The hypotheses are: (1) the strictly feasible set \(\Omega_1 := \{x\in \mathbb{R}^n: g(x)=0, h(x)<0\}\) is nonempty and bounded, (2) a regularity condition from which it may be concluded that \(\Omega_1\) is a smooth manifold, (3) a ``normal cone condition'' which generalizes the condition of convexity of \(\Omega_1\). The ``combined homotopy'' which the authors propose has three components. The first component is a convex combination of a linear map involving a feasible starting point together with the Lagrangian. The second component consists merely of the equality constraints, and the third component is a ``Newton homotopy'' involving the inequality constraints, making the combined method an interior point method. The parametrized Sard's theorem is applied to show that the combined homotopy method is globally convergent with probability one. A pair of numerical examples are given.