Locally, this looks a lot like the evaluation of a 2D convolution filter. Is there a way to represent a 3D geometry to evaluate these computations efficiently with a convolution filter?
skim
@clairezb It's the weighting for the midpoints as defined by the method, which was somewhat-arbitrarily defined by aesthetics.
Why are (m1+m2+m3+m4) multiplied by 2?