Fixing and extending some recent results on the ADMM algorithm

Area of Science:

• Optimization Theory
• Numerical Analysis
• Convex Analysis

Background:

• The study builds upon existing proximal alternating direction method of multipliers (ADMM) algorithms.
• Addresses limitations in handling specific types of convex optimization problems.

Purpose of the Study:

• To extend the convergence analysis of proximal ADMM algorithms.
• To formulate a new ADMM variant for problems with additional smooth functions.
• To investigate the use of variable metrics within the ADMM framework.

Main Methods:

• Leveraging techniques from Shefi and Teboulle (2014) for convergence analysis.
• Developing a modified ADMM algorithm incorporating a smooth function and its gradient.
• Implementing variable metric strategies within iterative steps.
• Conducting analysis in infinite-dimensional Hilbert spaces.

Main Results:

• Established convergence properties for the enhanced ADMM algorithms.
• Demonstrated the applicability of the new variant to a broader class of convex problems.
• Validated the effectiveness of variable metric approaches in this context.

Conclusions:

• The proposed algorithmic extensions enhance the capabilities of ADMM for complex convex optimization.
• The convergence analysis provides theoretical guarantees for the improved methods.
• This work contributes to the advancement of optimization techniques in infinite-dimensional settings.