1: \begin{abstract}

2: We propose a novel approach which employs random sampling to generate

3: an accurate non-uniform mesh for numerically solving Partial Differential

4: Equation Boundary Value Problems (PDE-BVP's). From a uniform probability

5: distribution $\mathcal{U}$ over a 1D domain, we sample $M$ discretizations

6: of size $N$ where $M\gg N$. The statistical moments of the solutions

7: to a given BVP on each of the $M$ ultra-sparse meshes provide insight

8: into identifying highly accurate non-uniform meshes. Essentially,

9: we use the pointwise mean and variance of the coarse-grid solutions

10: to construct a mapping $Q(x)$ from uniformly to non-uniformly spaced

11: mesh-points. The error convergence properties of the approximate solution

12: to the PDE-BVP on the non-uniform mesh are superior to a uniform mesh

13: for a certain class of BVP's. In particular, the method works well

14: for BVP's with locally non-smooth solutions. We present a framework

15: for studying the sampled sparse-mesh solutions and provide numerical

16: evidence for the utility of this approach as applied to a set of example

17: BVP's. We conclude with a discussion of how the near-perfect paralellizability

18: of our approach suggests that these strategies have the potential

19: for highly efficient utilization of massively parallel multi-core

20: technologies such as General Purpose Graphics Processing Units (GPGPU's).

21: We believe that the proposed algorithm is beyond embarrassingly parallel;

22: implementing it on anything but a massively multi-core architecture

23: would be scandalous.\end{abstract}

24: