Augmented Lagrangian Optimization under Fixed-Point Arithmetic

摘要: In this paper, we propose an inexact Augmented Lagrangian Method (ALM) for the optimization of convex and nonsmooth objective functions subject to linear equality constraints box where errors are due fixed-point data. To prevent data overflow also introduce a projection operation in multiplier update. We analyze theoretically proposed algorithm provide convergence rate results bounds on accuracy optimal solution. Since iterative methods often needed solve primal subproblem ALM, early stopping criterion that is simple implement embedded platforms, can be used problems not strongly convex, guarantees precision best our knowledge, first ALM handle non-smooth problems, overflow, efficiently systematically utilize solvers Numerical simulation studies utility maximization problem presented illustrate method.