My questions come from the pdf https://www.math.univ-toulouse.fr/~bordenave/coursSRG.pdf that I am reading. More specifically, on page 6, the author computes the spectral measure for finite graphs, and in particular for $C_n$ a cycle of length $n$.

If $G = (V, E)$ is a finite graph, $|V| = n$, the author defines the spectral measure of $G$ as

$$\mu_G = \frac{1}{n} \sum_{i = 1}^{n} \delta_{\lambda_i}$$

where $(\lambda_1, \ldots, \lambda_n)$ are the eigenvalues (counting multiplicity) of $A$, the adjacency operator.

Let $C_n$ be a cycle of length $n$. The adjacency operator can be written as $A = B + B^\ast$ where $B$ is the permutation matrix of a cycle of length $n$. Since $BB^\ast = B^\ast B = I$, the eigenvalues of $B$ are the roots of unity and the eigenvalues of $A$ are $\lambda_k = 2 \cos(2\pi k /n)$, $1 \le k \le n$. We get

$$\mu_{C_n} = \frac{1}{n} \sum_{i = 1}^{n} \delta_{2 \cos(2 \pi k / n)}$$

As $n$ goes to infinity, $\mu_{C_n}$ converges weakly to a arcsine distribution $\nu$ with density on $[-2,2]$ given by

$$d \nu(x) = \frac{1}{\pi \sqrt{4 - x^2}} {\large\chi}_{|x|\leq 2} dx,$$

($\nu$ is the law of $2 \cos(\pi U)$ with $U$ uniform on [0,1]).


I am not able to prove the weak convergence asserted by the author. Is someone able to prove it/elaborate on why this is true?


Let $X_n$ be the discrete random variable that attains the value $k/n$ with probability $1/n$, for $k=1,\dotsc,n$. The sequence $(X_n)_n$ converges in distribution to $U\sim\mathrm{Uniform}(0,1)$.

Consider the continuous function $g(x)=2\cos(2\pi x)$. By the properties of convergence in distribution, namely the continuous mapping theorem, the variables $g(X_n)$ converge in distribution to $g(U)$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.