On \(L_ p\)-norms of multivariate density estimators (Q1184216)

With \(n\) independent, identically distributed bivariate random vectors with distribution function \(F\) and density function \(f\), the authors consider kernel density estimators defined by \[ \bigl(nh^ 2(n)\bigr)^{-1}\sum_{1\leq i\leq n} K\bigl((t-X_ i)/h(n)\bigr), \] where \(K\) is a bivariate function satisfying certain regularity conditions and \(h(n)\) is a sequence of positive numbers. The paper considers the asymptotic properties of \[ I_ n(p)=\int^ \infty_{- \infty}\int^ \infty_{-\infty}| f_ n(t)-f(t)|^ pd\mu(t), \] where \(\mu\) is a weight function, the most popular choices of \(p\) are 1 and 2, and \(EI_ n(2)\) is the mean integrated square error. Having defined the asymptotic expected value of an appropriate function of \(I_ n(p)\), the main theorem in the paper is that for \(1\leq p<\infty\) a suitably normalized function of \(I_ n(p)\) has a limiting normal distribution. In the proof the author uses Poisson approximations of bivariate empirical processes and proves central limit theorems for \(L_ p\)-norms of kernel-transformed bivariate Poisson processes.