Attempting to compute the kslog numerically (i.e., Kneser's construction)

[jaydfox]

Anyway, to demonstrate the increased accuracy empirically, note the red lines. These correspond to the 105, 315, and 945 point systems, using 10-interval interpolation polynomials (11-points, with common endpoints).

Also, if you're wondering how I divide a 105-interval region into 10-interval sub-regions, note that the end regions are 7.5 intervals wide. So I would have an interpolating polynomial from 0 to 15/210, using the points 1/210, 3/210, 5/210, 7/210, 9/210, 11/210, 13/210, and 15/210. And thus on the other end of the interval as well (reflected, more or less).

As I start posting details and code, hopefully this will begin to make more sense.