1: \begin{abstract}

2: We consider nonparametric estimation of the mean and covariance

3: functions for functional/longitudinal data. Strong uniform convergence

4: rates are developed for estimators that are local-linear smooth\-ers. Our

5: results are obtained in a unified framework in which the number of

6: observations within each curve/cluster can be of any rate relative to

7: the sample size. We show that the convergence rates for the procedures

8: depend on both the number of sample curves and the number of

9: observations on each curve. For sparse functional data, these rates are

10: equivalent to the optimal rates in nonparametric regression. For dense

11: functional data, root-$n$ rates of convergence can be achieved with

12: proper choices of bandwidths. We further derive almost sure rates of

13: convergence for principal component analysis using the estimated

14: covariance function. The results are illustrated with simulation

15: studies.

16: \end{abstract}