1: \begin{abstract}

2: Network analysis is becoming one of the most active research areas

3: in statistics. Significant advances have been made recently on

4: developing theories, methodologies and algorithms for analyzing

5: networks. However, there has been little fundamental study on optimal

6: estimation. In this paper, we establish optimal rate of convergence for

7: graphon estimation. For the stochastic block model with $k$ clusters,

8: we show that the optimal rate under the mean squared error is

9: $n^{-1}\log k+k^2/n^2$. The minimax upper bound improves the existing

10: results in literature through a technique of solving a quadratic

11: equation. When $k\leq\sqrt{n\log n}$, as the number of the cluster

12: $k$ grows, the minimax rate grows slowly with only a logarithmic order

13: $n^{-1}\log k$. A key step to establish the lower bound is to construct

14: a novel subset of the parameter space and then apply Fano's lemma, from

15: which we see a clear distinction of the nonparametric graphon

16: estimation problem from classical nonparametric regression, due to the

17: lack of identifiability of the order of nodes in exchangeable random

18: graph models. As an immediate application, we consider nonparametric

19: graphon estimation in a H\"{o}lder class with smoothness $\alpha$.

20: When the smoothness $\alpha\geq1$, the optimal rate of convergence is

21: $n^{-1}\log n$, independent of $\alpha$, while for $\alpha\in(0,1)$,

22: the rate is $n^{-2\alpha/(\alpha+1)}$, which is, to our surprise,

23: identical to the classical nonparametric rate.

24: \end{abstract}