1: \begin{abstract}

2: Processing streaming data as they arrive is often necessary for high dimensional data analysis. In this paper,

3: we analyse the convergence of a subspace online PCA iteration, as a  followup of the recent work of

4: \LWLZ~[\emph{Math. Program., Ser. B}, DOI 10.1007/s10107-017-1182-z] who considered

5: the case for the most significant principal component only, i.e., a single vector. Under the sub-Gaussian

6: assumption, we obtain a finite-sample error bound that closely matches the minimax

7: information lower bound of Vu and Lei [\emph{Ann. Statist.\/}~\textbf{41}:6 (2013), 2905--2947].

8: \end{abstract}

9: