A decision process over variables and their number: Two-phase optimality conditions (Q792227): Difference between revisions

The authors are concerned with the following optimization problem (P): minimize \(f_ n(\bar x_ n,n)=g_ n(\bar x_ n)+h(n)\) where \[ g_ n(\bar x_ n)=\alpha \sum^{n-1}_{i=0}(\int^{x_{i+1}}_{x_ i}b(t)dt)dx \] over the integer variable n and \(\bar x_ n=(x_ 1,...,x_{n-1})\), \(x_ 0\) and \(x_ n\) are known boundary conditions and h(n) is increasing unbounded and weakly convex. Problems similar to (P) come out from several areas such as construction of minimal length functional equations, inventory theory, location theory, production problems. To solve (P) the authors consider a two phases strategy: minimize first \(f_ n(\bar x_ n,n)\) over \(\bar x_ n\) for a fixed n and then over n. They prove for (P) two main results: (i) if b(x) is such that \(\nabla g_ n(\bar x_ n^*)=0\) is sufficient for \(\bar x_ n^*\) to minimize \(g_ n(\bar x_ n)\) for every \((n,x_ 0,x_ n)\), then, the first order necessary condition for finding \(n^*\) is also sufficient; (ii) if b(x) is increasing and concave then the necessary first order condition \(\nabla g_ n(\bar x_ n^*)=0\) for \(\bar x_ n^*\) to minimize \(g_ n(\bar x_ n)\) for every \((n,x_ 0,x_ n)\) are also sufficient.