Distribution of levels in high-dimensional random landscapes (Q2428053)

For \(n\in\mathbb N\), let \(T_n\) be a finite collection of subsets \(t\) of cardinality \(|t|=n\) of a countable index set \(E\), \(V\) a random variable and \(\xi_e\), \(e\in E\), independent copies of a random variable \(\xi\) with expectation \(\mathbb E[\xi]=0\) and \(\mathbb E[\xi^2]=1\). Define the random field \(\mathbb X_n(t)=\frac{1}{\sqrt{n}}\sum_{e\in t} \xi_e\), and denote \(F_n(z)\),\(z\in\mathbb R\), its empirical distribution function and \(s_n^2\) the variance of \(\sum_{t\in T_n} \mathbb X_n(t)\). When \(n\to\infty\), assuming \(\frac 1{s_n} \sum_{t\in T_n}\mathbb X_n(t) \to V\) in distribution (together with a Lindeberg-type condition), the author prove the weak convergence \(\{\frac{|T_n|}{s_n} (F_n(z)- \mathbb E[F_n(z)]), z\in\mathbb R\}\to \{ -\frac{e^{-\frac{z^2}{2}}}{\sqrt{2\pi}} V, z\in\mathbb R\}\) for the finite-dimensional distributions. The result is then specialized to six specific random landscapes (cost of assignements, weights of Hamiltonian cycles and of spanning trees, locations of particles in branching random walk, energies of directed polymers and of spin configurations).