No label defined (Q3973903)

[For the entire collection see Zbl 0722.00032.] Let \(X_ n\), \(n\geq 1\) be i.i.d. r.v.s on \((-\infty,\infty)\) with distribution \(F\) having positive and continuous density \(f\) on a given non-void interval \(\langle A,B\rangle\), and let \[ f_ n(x)=(n\lambda_ n)^{-1}\sum^ n_{i=1} K((x-X_ i)/\lambda_ n) \] be the kernel estimate of \(f\) based on a sequence \(\{\lambda_ n, n\geq 1\}\) of positive smoothing parameters with the kernel \(K\) satisfying certain additional conditions. Moreover, let \[ E f_ n(x)=\int^ \infty_{- \infty}\lambda^{-1}_ n K((x-t)/\lambda_ n)dF(t). \] The author considers the sequence \[ \sup_{A<C\leq x\leq D < B}\left((f_ n(x)-E f_ n(x))/\sqrt{f(x)}\right)(n\lambda_ n/2 \hbox{log}(1/\lambda_ n))^{1/2} ,\qquad n\geq 1,\leqno(*) \] where \(\lambda_ n=\lambda(n;x,X_ 1,\dots,X_ n)\) are data-driven random bandwidths, and under suitable assumptions proves two limit theorems concerning convergence a.s. and in probability of (*) to the limit \((\int^ \infty_{-\infty}K^ 2(u)du)^{1/2}\).