1: \begin{abstract}

2: Semidefinite programs (SDPs) are powerful theoretical tools that have

3: been studied for over two decades, but their practical use remains

4: limited due to computational difficulties in solving large-scale,

5: realistic-sized problems. In this paper, we describe a modified interior-point

6: method for the efficient solution of large-and-sparse low-rank SDPs,

7: which finds applications in graph theory, approximation theory, control

8: theory, sum-of-squares, etc. Given that the problem data is large-and-sparse,

9: conjugate gradients (CG) can be used to avoid forming, storing, and

10: factoring the large and fully-dense interior-point Hessian matrix,

11: but the resulting convergence rate is usually slow due to ill-conditioning.

12: Our central insight is that, for a rank-$k$, size-$n$ SDP, the Hessian

13: matrix is ill-conditioned only due to a rank-$nk$ perturbation, which

14: can be explicitly computed using a size-$n$ eigendecomposition. We

15: construct a preconditioner to ``correct'' the low-rank perturbation,

16: thereby allowing preconditioned CG to solve the Hessian equation in

17: a few tens of iterations. This modification is incorporated within

18: SeDuMi, and used to reduce the solution time and memory requirements

19: of large-scale matrix-completion problems by several orders of magnitude.

20: \end{abstract}