交互乗数法
A Method of Multipliers with Alternating Constraints for Nonlinear Optimization Problems

概要

A new method of multipliers is proposed to solve constrained minimization problems which consist of equality and inequality constraints. The method solves a sequence of subproblems whose objective functions are an augmented Lagrangian function. The distinguishing feature of this method is that it allows the augmented Lagrangian function and its minimization problem to alternate the constraints at each iteration, that is some constraints are included in the augmented Lagrangian function, and the others remain in the subproblems. By alternating constraints, this method able to get more accurate estimated Lagrange multipliers by exploiting Karush-Kuhn-Tucker (KKT) points of the subproblems, and consequently it will converge more efficient and steady.

産業界への展開例・適用分野

Some special structures of the subproblems are discussed for efficiency. Numerical experiments are presented where this method were applied with a proximal gradient method to solve the subproblems of the large scale convex programming problems with linear constraints.

研究者

氏名	専攻	研究室	役職/学年
Hassan Siti Nor Habibah bin	数理工学専攻	System Optimization	博士2回生

---