1: \begin{abstract}

2: 	In this paper we analyze the behavior of the Oja's algorithm for online/streaming principal component subspace estimation.

3: 	It is proved that with high probability it performs an efficient, gap-free, global convergence rate to approximate an principal component subspace for any sub-Gaussian distribution.

4: 	Moreover, it is the first time to show that the convergence rate, namely the upper bound of the approximation, exactly matches the lower bound of an approximation obtained by the offline/classical PCA up to a constant factor.

5: \end{abstract}

6: