1: \begin{abstract}

2: We consider the sequence acceleration problem for the alternating

3: direction method-of-multipliers (ADMM) applied to a class of equality-constrained

4: problems with strongly convex quadratic objectives, which frequently

5: arise as the Newton subproblem of interior-point methods. Within this

6: context, the ADMM update equations are linear, the iterates are confined

7: within a Krylov subspace, and the General Minimum RESidual (GMRES)

8: algorithm is optimal in its ability to accelerate convergence. The

9: basic ADMM method solves a $\kappa$-conditioned problem in $O(\sqrt{\kappa})$

10: iterations. We give theoretical justification and numerical evidence

11: that the GMRES-accelerated variant consistently solves the same problem

12: in $O(\kappa^{1/4})$ iterations, for an order-of-magnitude reduction

13: in iterations, despite a worst-case bound of $O(\sqrt{\kappa})$ iterations.

14: The method is shown to be competitive against standard preconditioned

15: Krylov subspace methods for saddle-point problems. The method is embedded

16: within SeDuMi, and used to solve many large-scale semidefinite programs

17: with error that decreases like $O(1/k^{2})$, instead of $O(1/k)$,

18: where $k$ is the iteration index.

19: \end{abstract}