Primal-dual proximal algorithms for structured convex optimization: a unifying framework (Q2415201)

The authors introduce a novel primal-dual framework for solving structured convex optimization problems involving the sum of a Lipschitz-differentiable function and two nonsmooth functions, one of which composed with a linear continuous operator. The new framework is based on an asymmetric forward-backward-adjoint three-term splitting and it contains as special cases some known algorithms and some new primal-dual schemes. Linear convergence of the new method is achieved under four different regularity assumptions for the cost functions and for the class of problems with piecewise linear-quadratic cost functions. For the entire collection see [Zbl 1407.90006].