A Note on the Forward-Douglas--Rachford Splitting for Monotone Inclusion and Convex Optimization

TL;DR

This paper clarifies the structure of the forward-Douglas--Rachford splitting algorithm for monotone operators, extends it to multiple operators with preconditioning, and demonstrates its effectiveness in nonsmooth convex optimization.

Contribution

It provides a clear formulation of the true forward-Douglas--Rachford splitting algorithm and extends it to multiple operators with preconditioning, with experimental validation.

Findings

The algorithm effectively solves monotone inclusion problems.

Extension to multiple operators with preconditioning improves flexibility.

Experimental results show advantages in nonsmooth convex optimization.

Abstract

We shed light on the structure of the "three-operator" version of the forward-Douglas--Rachford splitting algorithm for finding a zero of a sum of maximally monotone operators $A + B + C$, where $B$ is cocoercive, involving only the computation of $B$ and of the resolvent of $A$ and of $C$, separately. We show that it is a straightforward extension of a fixed-point algorithm proposed by us as a generalization of the forward-backward splitting algorithm, initially designed for finding a zero of a sum of an arbitrary number of maximally monotone operators $\sum_{i=1}^n A_i + B$, where $B$ is cocoercive, involving only the computation of $B$ and of the resolvent of each $A_i$ separately. We argue that, the former is the "true" forward-Douglas--Rachford splitting algorithm, in contrast to the initial use of this designation in the literature. Then, we highlight the extension to an arbitrary number of maximally monotone operators in the splitting, $\sum_{i=1}^{n} A_i + B + C$, in a formulation admitting preconditioning operators. We finally demonstate experimentally its interest in the context of nonsmooth convex optimization.