Minimization of a Non-Separable Objective Function Subject to Disjoint Constraints

Abstract

We consider an important class of mathematical programs, in which the vector variable can be partitioned into two subvectors corresponding to independent constraint sets. Necessary and sufficient conditions for optimal solutions are developed, and two approaches for obtaining solutions are reviewed. We present an enumeration approach, reducing the problem to a finite number of subproblems, and show that duality makes the solution of many of the subproblems unnecessary. Next, we develop an alternating approach, wherein the problem is solved for one of the subvectors while the other is held constant, and then the subvector roles are reversed. This procedure is observed to converge to partial optimum solutions. A widely applicable subclass of problems includes a linear program in one of the subvectors. For this subclass a sufficient condition for local optimality is determined. The condition is easily testable and fails to hold, in many cases, only if a better solution is obtained. Also, this condition shows that partial optimum solutions are almost always local optima.

Efficient Spectrum Sensing for Cognitive Radio Networks via Joint Optimization of Sensing Threshold and Duration

“…Regarding minimizing of a biconvex function f : X × Y → R with a partial optimum (x * , y * ) ∈ X × Y , Wendell and Hurter [15] …”

Unsupervised Color Image Segmentation Using Compound Markov Random Field Model

“…These partial optimal solutions Z * and θ * are not global maxima, rather they are almost always local optimal solutions [8]. But with θ = θ * , the estimate z * is global optimal satisfying equation (2) and analogously for z = z * , θ * is global optimal satisfying equation(3).…”

“…In (1), z, θ could be viewed as a set of parameter of the given function P (Z = z/X = x, θ). For such kind of problems in deterministic framework, Wendell and Horter [8] have proposed an alternate approach that would yield suboptimal solutions instead of optimal solution. Their approach is based on splitting the variables followed by recursively estimating the parameters.…”

Maximally entangled mixed states for qubit-qutrit systems

“…5, a number of 24 optimization rounds were necessary to fulfill the adopted convergence (stop) criterion N P,n − N P,n−1 < 10 −12 (52) for all considered values of P ∈] The occurrence of these "subconverged points" is a direct consequence of the fact that ACS is not guaranteed to converge to a global optimum [31]. As a matter of fact, one cannot even be sure that the massive convergence to N (deg) X,P represents global optimum convergence.…”