1: \begin{abstract}

2: We consider nonparametric estimation of $L_2$, R\'{e}nyi-$\alpha$ and Tsallis-$\alpha$ divergences between continuous distributions.

3: Our approach is to construct estimators for particular integral functionals of two densities and translate them into divergence estimators.

4: For the integral functionals, our estimators are based on corrections of a preliminary plug-in estimator.

5: We show that these estimators achieve the parametric convergence rate of $n^{-1/2}$ when the densities' smoothness, $s$, are both at least $d/4$ where $d$ is the dimension.

6: We also derive minimax lower bounds for this problem which confirm that $s > d/4$ is necessary to achieve the $n^{-1/2}$ rate of convergence.

7: We validate our theoretical guarantees with a number of simulations.

8: %% We first unify these three estimation problems to that of estimating integral functionals of two densities and then construct several estimators for these quantities.

9: \end{abstract}

10: