A unified convergence theory is derived for a class of stationary iterative methods for solving linear equality constrained quadratic programs or saddle point problems.This class is constructed from essentially all po...A unified convergence theory is derived for a class of stationary iterative methods for solving linear equality constrained quadratic programs or saddle point problems.This class is constructed from essentially all possible splittings of the submatrix residing in the(1,1)-block of the augmented saddle point matrix that would produce non-expansive iterations.The classic augmented Lagrangian method and alternating direction method of multipliers are two special members of this class.展开更多
We establish local convergence results for a generic algorithmic framework for solving a wide class of equality constrained optimization problems.The framework is based on applying a splitting scheme to the augmented ...We establish local convergence results for a generic algorithmic framework for solving a wide class of equality constrained optimization problems.The framework is based on applying a splitting scheme to the augmented Lagrangian function that includes as a special case the well-known alternating direction method of multipliers(ADMM).Our local convergence analysis is free of the usual restrictions on ADMM-like methods,such as convexity,block separability or linearity of constraints.It offers a much-needed theoretical justification to the widespread practice of applying ADMM-like methods to nonconvex optimization problems.展开更多
基金This paper is a polished version of the Rice University technical report CAAMTR10-24which was a work supported in part by the National Natural Science Foundation(No.DMS-0811188)Office of Navy Research(No.N00014-08-1-1101).
文摘A unified convergence theory is derived for a class of stationary iterative methods for solving linear equality constrained quadratic programs or saddle point problems.This class is constructed from essentially all possible splittings of the submatrix residing in the(1,1)-block of the augmented saddle point matrix that would produce non-expansive iterations.The classic augmented Lagrangian method and alternating direction method of multipliers are two special members of this class.
基金This work was supported in part by Shenzhen Fundamental Research Fund(Nos.JCYJ-20170306141038939,KQJSCX-20170728162302784,ZDSYS-201707251409055)via the Shenzhen Research Institute of Big DataThe work of Jun-Feng Yang was supported by the National Natural Science Foundation of China(Nos.11771208,11922111,11671195).
文摘We establish local convergence results for a generic algorithmic framework for solving a wide class of equality constrained optimization problems.The framework is based on applying a splitting scheme to the augmented Lagrangian function that includes as a special case the well-known alternating direction method of multipliers(ADMM).Our local convergence analysis is free of the usual restrictions on ADMM-like methods,such as convexity,block separability or linearity of constraints.It offers a much-needed theoretical justification to the widespread practice of applying ADMM-like methods to nonconvex optimization problems.