1: \begin{abstract}

2: Eigenvectors and  eigenvalues of  discrete  graph  Laplacians are often used

3: for manifold learning and nonlinear dimensionality reduction. It was previously proved by Belkin and Niyogi \cite{CLEM_08}

4:  that the eigenvectors and eigenvalues of the graph Laplacian converge to the eigenfunctions

5: and eigenvalues of the Laplace-Beltrami operator of the manifold in the limit of infinitely many

6: data points sampled independently from the uniform distribution over the manifold.  Recently, we

7: introduced Point Integral method (PIM) \cite{LSS,SS-neumann} to solve elliptic equations and corresponding eigenvalue problem

8: on point clouds. We have established a unified

9: framework to approximate the elliptic differential operators on point clouds.

10: In this paper, we prove that the eigenvectors

11: and eigenvalues obtained by PIM

12: converge in the limit of infinitely many random samples independently from a distribution (not necessarily to be uniform distribution).

13: Moreover, one estimate of the rate of the convergence is also given.

14: \end{abstract}

15: