Chung--Smirnov property for perturbed empirical distribution functions (Q1209460): Difference between revisions

The distribution function (DF) \(F\) is estimated by using the integral of a sequence of kernel density estimators. It converges to the \(df\) of the unit mass zero. The convergence of the derived estimator \(\hat F_ n\) is well known. The author obtains new convergence criteria. Three conditions related to integrals of kernel functions are used for proving that \[ \limsup_{n\to\infty} (2n/\log\log (n))\|\hat F_ n-F\|=1; \] this is the Chung-Smirnov property. This result holds using an other set of hypothesis. New conditions on the kernel functions' integrals are included among the hypotheses. The conditions used for proving the two theorems, that provide the main results of the paper, are weaker than the usual ones, based on the proposals of \textit{B. B. Winter} [J. Appl. Probab. 16, 163-173 (1979; Zbl 0403.60033)].