HIERARCHIES OF HIGHER-ORDER KERNELS

Recent literature on functional estimation has shown the importance of kernels with vanishing moments although no general framework was given to build kernels of increasing order apart from some specific methods based on moment relationships. The purpose of the present paper is to develop such a framework and to show how to build higher order kernels with nice properties and to solve optimization problems about kernels. The proofs given here, unlike standard variational arguments, explain why some hierarchies of kernels do have optimality properties. Applications are given to functional estimation in a general context. In the last section special attention is paid to density estimates based on kernels of order (m, r), i.e., kernels of order r for estimation of derivatives of order m. Convergence theorems are easily derived from interpretation by means of projections in L2 spaces.