Abstract:Given i.i.d. observations uniformly distributed on a closed manifold $\mathcal{M}\subseteq \mathbb{R}^p$, we study the spectral properties of the associated empirical graph Laplacian based on a Gaussian kernel. Our main results are non-asymptotic error bounds, showing that the eigenvalues and eigenspaces of the empirical graph Laplacian are close to the eigenvalues and eigenspaces of the Laplace-Beltrami operator of $\mathcal{M}$. In our analysis, we connect the empirical graph Laplacian to kernel principal component analysis, and consider the heat kernel of $\mathcal{M}$ as reproducing kernel feature map. This leads to novel points of view and allows to leverage results for empirical covariance operators in infinite dimensions.

What problem does this paper attempt to address?

This paper aims to study the empirical spectral properties of the graph Laplacian matrix corresponding to independent and identically distributed (i.i.d.) observations uniformly distributed on a closed manifold $M \subseteq \mathbb{R}^p$. Specifically, the paper focuses on how to approximate the eigenvalues and eigenspaces of the Laplace - Beltrami operator of the manifold $M$ by the empirical graph Laplacian matrix constructed with the Gaussian kernel. The main results of the authors are non - asymptotic error bounds, which show that the eigenvalues and eigenspaces of the empirical graph Laplacian matrix are very close to those of the Laplace - Beltrami operator. ### Main problems solved in the paper 1. **Spectral approximation problem**: - Study how to approximate the spectral characteristics of the Laplace - Beltrami operator of the manifold $M$ by the empirical graph Laplacian matrix. - Provide non - asymptotic error bounds to prove the closeness between the eigenvalues and eigenspaces of the empirical graph Laplacian matrix and those of the Laplace - Beltrami operator. 2. **Heat Kernel Principal Component Analysis (Heat Kernel PCA)**: - Connect the empirical graph Laplacian matrix with Heat Kernel PCA, and provide a new functional analysis framework by using the heat kernel as the feature mapping in the Reproducing Kernel Hilbert Space (RKHS). - Use the results of the empirical covariance operator in infinite - dimensional spaces to provide a new perspective for the analysis of the empirical graph Laplacian matrix. 3. **Application of perturbation theory**: - Apply Hilbert space perturbation theory to study the stability of the eigenvalues and eigenspaces of the empirical graph Laplacian matrix. - Propose new perturbation bounds, especially the relative perturbation bounds for the null space, in order to obtain more refined error bounds. ### Key technical means - **Heat kernel and heat semigroup**: - Regard the Laplace - Beltrami operator as the generator of the heat semigroup, and use the heat kernel as the integral kernel to define the heat semigroup. - By restricting the domain of the heat semigroup to the Reproducing Kernel Hilbert Space (RKHS) induced by the heat kernel, interpret it as the covariance operator of random variables. - **Empirical covariance operator**: - Utilize the properties of the empirical covariance operator to control random errors through concentration inequalities and empirical process theory. - Apply classical perturbation theories (such as Weyl bounds and Davis - Kahan bounds) to derive error bounds for eigenvalues and eigenspaces. - **Advanced techniques**: - Use more advanced techniques, such as concentration inequalities, perturbation bounds, and eigenvalue calculations, to improve error bounds. - Especially when dealing with Heat Kernel PCA and the leading eigenvalues and their associated eigenspaces, more refined techniques are required. ### Conclusion By connecting the empirical graph Laplacian matrix with Heat Kernel PCA, the paper provides a new functional analysis framework and successfully solves the key problems of the empirical graph Laplacian matrix in approximating the spectral characteristics of the Laplace - Beltrami operator. These results are not only of great theoretical significance but also provide a solid theoretical basis for practical applications.

A kernel-based analysis of Laplacian Eigenmaps

An Error Bound for Eigenvalues of Graph Laplacian with Bounded Kernel Function

Convergence of Laplacian Spectra from Random Samples

Extreme eigenvalues of Laplacian random matrices with Gaussian entries

Laplacian Eigenmaps for Dimensionality Reduction and Data Representation

From Varadhan's Limit to Eigenmaps: A Guide to the Geometric Analysis behind Manifold Learning

Fast Spectrum Estimation of Some Kernel Matrices

An Infinite Dimensional Analysis of Kernel Principal Components

Locality preserving kernel on graphs

The geometry of kernelized spectral clustering

Error Analysis of Kernel/GP Methods for Nonlinear and Parametric PDEs

Extremal Eigenvalues of Random Kernel Matrices with Polynomial Scaling

A Note on Laplacian Eigenmaps

Spectrum of Laplacian matrices associated with large random elliptic matrices

Statistical analysis of random objects via metric measure Laplacians

Bergman kernels and eigenvalue estimate of $\bar{\partial}$-laplacian

Unbounded Laplacians on Graphs: Basic Spectral Properties and the Heat Equation

Fluctuation of the Largest Eigenvalue of a Kernel Matrix with application in Graphon-based Random Graphs

Random matrices with row constraints and eigenvalue distributions of graph Laplacians

On the Eigenvectors of P-Laplacian