Abstract:We study the asymptotics of the expected Wasserstein distance between the empirical measure of a Point Process and the background volume form. The main DPP studied is the harmonic ensemble, where we get the optimal rate of convergence for homogeneous manifolds of dimension $d\geq 3$, and for two-point homogeneous manifolds. We also discuss some variations of this process on the torus. Regarding other point processes, we find the optimal rate for the spherical ensemble and the zeros of Gaussian Analytic Functions.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is how to quantify and optimize the convergence of the Wasserstein distance between the empirical measure of point processes (especially harmonic ensembles) and the background volume form. Specifically, the author has studied how the empirical measures of Determinantal Point Processes (DPPs) approximate the background measures on different dimensions and manifolds, and has given the optimal convergence rate.
### Main Research Contents
1. **Optimal Convergence Rates of Harmonic Ensembles**:
- For homogeneous manifolds with dimension \(d \geq 3\), the author has obtained the optimal convergence rate of the Wasserstein distance between the empirical measure of harmonic ensembles and the background measure.
- For two - point homogeneous manifolds, the author has also obtained similar optimal results, especially for the case where the dimension \(d \geq 2\).
2. **Research on Other Point Processes**:
- The author has also studied the convergence of spherical ensembles and the zeros of Gaussian analytic functions, and has given the optimal convergence rate of the Wasserstein distance between the empirical measures of these point processes and the background measure.
3. **Variants on the Torus**:
- Discussed the changes of harmonic ensembles on the torus \(T^2\) and proved the optimal convergence rate.
### Formula Summary
- **Definition of Wasserstein Distance**:
\[
W_p(\mu, \nu) = \left( \inf_{\pi \in \Pi(\mu, \nu)} \int_{M \times M} d(x, y)^p \, d\pi(x, y) \right)^{1/p}
\]
where \(\Pi(\mu, \nu)\) is all probability measures with \(\mu\) and \(\nu\) as marginal distributions.
- **Results of the Main Theorem**:
- For finite - dimensional projection DPPs with dimension \(d \geq 3\):
\[
\mathbb{E} W_2\left( \frac{1}{N} \sum_{n = 1}^N \delta_{x_n}, \text{Vol} \right) \lesssim N^{-1/d}
\]
- For dimension \(d = 2\):
\[
\mathbb{E} W_2\left( \frac{1}{N} \sum_{n = 1}^N \delta_{x_n}, \text{Vol} \right) \lesssim \sqrt{\log N} N^{-1/2}
\]
- **Results on Two - Point Homogeneous Manifolds**:
\[
\mathbb{E} W_2\left( \frac{1}{N} \sum_{n = 1}^N \delta_{x_n}, \text{Vol} \right) \lesssim N^{-1/d}
\]
### Method Overview
The author used a "smoothing inequality", similar to the Berry - Esseen - type inequality, to deal with the Wasserstein distance \(W_2\). Through this method, the author was able to transform the problem into the estimation of certain integrals and variances, thus obtaining the optimal results of convergence.
### Conclusion
Through strict mathematical derivation and analysis, this paper has proved the optimal convergence rates of the Wasserstein distance between the empirical measures of harmonic ensembles and other related point processes and the background measure on different types of manifolds. This not only deepens our understanding of these point processes but also provides an important theoretical basis for future research.