The augmented Lagrangian method can approximately solve convex optimization with least constraint violation (Q6102759)

The authors propose a procedure appropriate for optimization problems, in which it is not known whether the set of feasible solutions in non-empty. They replace the original optimization problem with an optimization problem with shifted right-hand sides of the constraints and solve the shifted problem under certain restriction of the shift. They obtain as a result an optimal solution, which is in general infeasible for the original (non-shifted) optimization problem, but the restriction of the shift makes possible to ensure minimal violation of the constraints of the original optimization problem. The proposed methodology is explained on convex optimization problems \[ f(x) \rightarrow \min \text{ s. t. }g(x) = y,\;\; y\in K, \] where \(f: R^n \rightarrow R\), \(g: R^n \rightarrow Y\), \(y\in K\), \(K\) is a closed convex set in a finite-dimensional space. Necessary and sufficient optimality conditions using augmented Lagrangian are derived. Theoretical results are illustrated by a numerical example.