1: \begin{abstract}

2: Spectral clustering is a technique that clusters elements using the top

3: few eigenvectors of their (possibly normalized) similarity matrix.

4: The quality of spectral clustering is closely tied to the convergence

5: properties of these principal eigenvectors.

6: This rate of convergence has been shown to be identical for both the

7: normalized and unnormalized variants in recent random matrix theory literature.

8: However, normalization for spectral clustering is commonly believed to

9: be beneficial [\textit{Stat. Comput.} \textbf{17} (2007)

10: 395--416]. Indeed, our experiments show that normalization improves

11: prediction accuracy.

12: In this paper, for the popular stochastic blockmodel, %we

13: %theoretically show that under spectral embedding, normalization

14: %shrinks the variance of points in a class by a constant fraction.

15: we theoretically show that normalization shrinks the spread of points

16: in a class by a constant fraction under a broad parameter regime.

17: As a byproduct of our work, we also obtain sharp deviation bounds of

18: empirical principal eigenvalues of graphs generated from a stochastic

19: blockmodel.

20: \end{abstract}