Multilevel kernel approximation of log-Euclidean manifold-valued functions (Q7422385)

We study kernel approximation for manifold-valued functions with global logarithm onto a Hilbert space, approximating in logarithmic coordinates to keep the approximant on the manifold while transferring error estimates; we also prove metric comparisons between log-Euclidean and affine-invariant distances and obtain multilevel convergence with sparse systems.