Precise optimization using range arithmetic (Q1344345): Difference between revisions

The author considers the search algorithm for local minima of smooth unconstrained functions given by \textit{R. Fletcher} [Practical methods of optimization, Vol. 1: Unconstrained optimization (1980; Zbl 0439.93001), Chapter 5]: (\(k\)th iteration) (i) given \(x^k\) and \(\nu^k\), calculate \(g^k\); if \(g^k = 0\), then exit; (ii) calculate \(G^k\); while \(G^k + \nu^k I\) is not positive definite, increase \(\nu^k\); (iii) solve \((G^k + \nu^k I) \delta^k = -g^k\) for \(\delta^k\); (iv) evaluate \(r^k = \{f(x^k) - f(x^k + \delta^k)\} /\{q^k(0) - q^k(\delta^k)\}\); (v) if \(r^k > 0.75\), then reduce \(\nu^k\); if \(r^k < 0.25\), then increase \(\nu^k\); set \(\nu^{k+1} = \nu^k\); (vi) if \(r^k \leq 0\), then \(x^{k+1} = x^k\); otherwise \(x^{k + 1} = x^k + \delta^k\). The author shows how this algorithm can be adapted to range arithmetic in order to guarantee convergence and generate correct results. Finally, he provides some numerical examples.