1: \begin{abstract}

2:  The discretization of the double-layer potential integral equation

3:  for the interior Dirichlet Laplace problem in a domain with smooth

4:  boundary results in a linear system that has a bounded condition

5:  number. Thus, the number of iterations required for the convergence

6:  of a Krylov method is, asymptotically, independent of the

7:  discretization size $N$. Using the Fast Multipole Method (FMM) to

8:  accelerate the matrix-vector products, we obtain an optimal

9:  $\bigO(N)$ solver. In practice, however, when the geometry is

10:  complicated, the number of Krylov iterations can be quite large---to

11:  the extend that necessitates the use of preconditioning.

12: 

13:   We summarize the different methodologies that have appeared in the

14:   literature (single-grid, multigrid, approximate sparse inverses) and

15:   we propose a new class of preconditioners based on an FMM-based

16:   spatial decomposition of the double-layer operator.  We present an

17:   experimental study in which we compare the different approaches and

18:   we discuss the merits and shortcomings of our approach.  Our method

19:   can be easily extended to other second-kind integral equations with

20:   non-oscillatory kernels in two and three dimensions.

21: \end{abstract}

22: