1: \begin{abstract}

2: Principal component analysis (PCA) is one of the most commonly used

3: statistical procedures with a wide range of applications. This paper

4: considers both minimax and adaptive estimation of the principal

5: subspace in the high dimensional setting. Under mild technical

6: conditions, we first establish the optimal rates of convergence for

7: estimating the principal subspace which are sharp with respect to all

8: the parameters, thus providing a complete characterization of the

9: difficulty of the estimation problem in term of the convergence rate.

10: The lower bound is obtained by calculating the local metric entropy and

11: an application of Fano's lemma. The rate optimal estimator is

12: constructed using aggregation, which, however, might not be

13: computationally feasible.

14: 

15: We then introduce an adaptive procedure for estimating the principal

16: subspace which is fully data driven and can be computed efficiently. It

17: is shown that the estimator attains the optimal rates of convergence

18: simultaneously over a large collection of the parameter spaces. A key

19: idea in our construction is a reduction scheme which reduces the sparse

20: PCA problem to a high-dimensional multivariate regression problem. This

21: method is potentially also useful for other related problems.

22: \end{abstract}