The Chebyshev nodes prevent Runge's phenomen when determining polynomials. For the interval [0, 1] the Chebyshev nodes of second type can be determined by (1 + math.cos(math.pi * (1 + (i-1) / (n-1)))) / 2 with i=1..n.
Is there a multidimensonal version which distributes the points "optimally" on a multidimensional simplex?
I've been looking into numpy's chebyshev module, but didn't get the connection to the nodes.