1: \begin{abstract}

2: % We consider a family of general block coordinate descent (BCD) methods in which a single block of variables is updated at each iteration. Such BCD type methods are well suited  for large scale nonsmooth constrained convex optimization involving multiple block variables since they solve an easy low dimensional problem at each iteration, often in close form.   Despite their popularity, the characterization of the iteration complexity for the BCD-type algorithms has remained incomplete to date, especially for multi-block nonsmooth problems and when the block variables are updated in a deterministic fashion, for example by using the classic Guass-Seidel (G-S), Gauss-Southwell (G-So), or Essentially Cyclic (E-C) rules.

3: 

4:  In this paper, we provide a unified iteration complexity analysis for a family of general block coordinate descent (BCD) methods, covering popular methods such as the block coordinate gradient descent (BCGD) and the block coordinate proximal gradient (BCPG), under various different coordinate update rules. We unify these algorithms under the so-called Block Successive Upper-bound Minimization (BSUM) framework, and show that for a broad class of multi-block nonsmooth convex problems, all algorithms covered by the BSUM framework achieve a global sublinear iteration complexity of ${\cal{O}}(1/r)$, where $r$ is the iteration index. { Moreover, for the case of block coordinate minimization (BCM) where each block is minimized exactly, we establish the sublinear convergence rate of $O(1/r)$ without per block strong convexity assumption. Further, we show that when there are only two blocks of variables, a special BSUM algorithm with Gauss-Seidel rule can be accelerated to achieve an improved rate of ${\cal{O}}(1/r^2)$.}

5: \end{abstract}