Abstract:The circular unitary ensemble and its generalizations concern a random matrix from a compact classical group $\mathrm{U}(N)$, $\mathrm{SU}(N)$, $\mathrm{O}(N)$, $\mathrm{SO}(N)$ or $\mathrm{USp}(N)$ distributed according to the Haar measure. The eigenvalues are known to be very evenly distributed on the unit circle. In this paper, we study the distance from the empirical measure of the eigenvalues to uniformity in the quadratic Wasserstein metric $W_2$. After finding the exact value of the expected value and the variance, we deduce a limit law for the square of the Wasserstein distance. We reformulate our results in terms of the $L^2$ average of the number of eigenvalues in circular arcs, and also in terms of the characteristic polynomial of the matrix on the unit circle.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the distance between the empirical measure of eigenvalues of random matrices and the uniform distribution under the Wasserstein metric. Specifically, the author studies the distribution of eigenvalues on the unit circle of random matrices from the compact classical groups \(U(N)\), \(SU(N)\), \(O(N)\), \(SO(N)\) or \(USp(N)\), and focuses on the quadratic Wasserstein distance \(W_2\) between the empirical measure of these eigenvalues and the uniform distribution on the unit circle.
The main contributions of the paper include:
1. **Exact calculation of expectation and variance**: The author derives the exact asymptotic expressions for the expectation and variance of \(W_2^2(\mu_A, \lambda_T)\).
2. **Limit theorem**: The author proves the limit distribution of \(W_2^2(\mu_A, \lambda_T)\) after appropriate centering and scaling.
### Specific problem description
In random matrix theory, a central problem is to study the empirical spectral measure of random matrix models. For large matrices, if the elements of the matrix are independently and identically distributed, under very general assumptions, the empirical measure of eigenvalues will converge to Wigner's semicircle law (in the case of Hermitian matrices) or the uniform measure on the complex unit disk (in the case of non - Hermitian matrices) as the matrix dimension increases. In recent years, estimating the rate of this convergence under the Wasserstein metric has attracted wide attention.
This paper mainly studies Dyson's circular unitary ensemble and some of its generalizations. Specifically, consider matrices randomly and uniformly sampled from the compact classical groups \(U(N)\), \(SU(N)\), \(O(N)\), \(SO(N)\) or \(USp(N)\). The \(N\) eigenvalues of these matrices are located on the unit circle and are very evenly distributed.
### Main results
1. **Exact asymptotic expressions for expectation and variance**:
- For \(G = U(N)\), \(SU(N)\), \(SO(2N + 1)\), \(O(2N + 1)\), \(SO(2N)\), \(O^-(2N+2)\) or \(USp(2N)\), the expectation and variance of the quadratic Wasserstein distance \(W_2^2(\mu_A, \lambda_T)\) between the empirical measure \(\mu_A\) of eigenvalues of the random matrix \(A\) and the uniform measure \(\lambda_T\) on the unit circle are respectively:
\[
\mathbb{E}[W_2^2(\mu_A, \lambda_T)]=\frac{2\log N_0}{N_0^2}+\frac{c_G}{N_0^2}+O\left(\frac{1}{N_0^3}\right),
\]
\[
\text{Var}[W_2^2(\mu_A, \lambda_T)]=\frac{\sigma_G}{N_0^4}+O\left(\frac{1}{N_0^5}\right),
\]
where \(N_0\) is the number of non - trivial eigenvalues, and the constants \(c_G\) and \(\sigma_G\) are shown in Table 1.
2. **Limit theorem**:
- The author proves that \(N_0^2W_2^2(\mu_A, \lambda_T)- 2\log N_0 - c_G\) converges in distribution to a certain random variable \(\xi_G\) as \(N\rightarrow\infty\), whose definition is shown in Table 2 and the characteristic function is shown in Table 3.
### Method
The author's method depends on two key facts:
1. The non - trivial eigenvalues form a deterministic point process.
2. Any Borel probability measure \(\mu\) and the uniform measure on the unit circle.