Calculating discrepancies (esp. L2) and searching for optimized nets

[ParadaCarleton]

There are fast discrepancy calculations that can be used to estimate the quality of a low-discrepancy point set, see Section 10.2 (page 167) for example. We could then try optimizing point sets to minimize these discrepancies; L2 is both easy to calculate and closely related to the mean squared error when the function being integrated is modeled as a Gaussian process.