Samplet compression for conditionally positive definite kernels and universal Kriging (Q7284988)

We introduce a fast numerical method for conditionally positive definite (CPD) kernel approximation using samplets. By exploiting the vanishing moment property of samplets and their associated scaling distributions, we reduce an indefinite saddle-point system to a linear and triangular system that can be solved efficiently by sparse Cholesky factorization with O(N log N) cost.