Before 1900, E. De Forest studied data smoothing by kernels that change sign. This leads naturally to limit theorems involving higher order differential operators. We will discuss similarities and differences with the case of probability distributions on $\mathbb Z$ and on $\mathbb Z^d$.