1: \begin{abstract}

2: This paper is concerned with

3: approximating the dominant left singular vector space of a real matrix $\bA$ of arbitrary dimension,

4: from block Krylov spaces generated by the matrix $\bA\bA^T$ and the block vector $\bA\bX$.

5: Two classes of results are presented. First are bounds on the distance, in the two and Frobenius norms,

6: between the Krylov space and the target space. The distance is expressed in terms

7: of principal angles. Second are quality of approximation bounds, relative

8: to the best approximation in the Frobenius norm.

9: For starting guesses $\bX$ of full column-rank,

10: the bounds depend on the tangent of the principal angles between $\bX$ and the

11: dominant right singular vector space of $\bA$.

12: The results presented here form the structural foundation for

13: the analysis of randomized Krylov space methods. The innovative feature is a combination of

14: traditional Lanczos convergence analysis with optimal

15: approximations via least squares problems.

16: \end{abstract}

17: