1: \begin{abstract}

2: Convergence rates are established for an inexact accelerated

3: alternating direction method of multipliers (I-ADMM) for general 

4: separable convex optimization with a linear constraint. 

5: Both ergodic and non-ergodic iterates are analyzed.

6: Relative to the iteration number $k$, the convergence rate is $\C{O}(1/k)$

7: in a convex setting and $\C{O}(1/k^2)$ in a strongly convex setting.

8: When an error bound condition holds, the algorithm is 2-step linearly

9: convergent.

10: The I-ADMM is designed so that the accuracy of the inexact iteration

11: preserves the global convergence rates of the {\it exact} iteration,

12: leading to better numerical performance in the test problems.

13: \end{abstract}

14: