scientific article; zbMATH DE number 3876374

From MaRDI portal
Publication:3341648





Consider a sequence of Gibbs measures with mean field interaction \[ d\nu_ n=Z_ n^{-1}\exp (n^{-1}\sum^{n}_{i,j=1}V(x_ i,x_ j))\otimes^{\infty}_{i=1}\mu (dx_ i). \] It is known from the theory of large deviations that \(\nu_ n\) converges weakly to the set \(P_ 0\) of those probability measures minimizing the free energy \(F(R)=- \int V(x,y)R(dx)R(dy)+\int\log (dR/d\mu)(x)\quad R(dx)\). The present paper gives interesting results in the case where \(P_ 0\) has more than one element. If the second-order Fréchet differential \(D^ 2F(R)\) is strictly positive definite for all \(R\in P_ 0\), then \(P_ 0\) is finite and \(\nu_ n\) converges weakly to \(\sum a_ kR_ k\) with \(a_ k\) proportional to \(\det (D^ 2F(R_ k))^{-1/2}\). If the spectrum of \(D^ 2F(R)\) contains zero for exactly one \(R_ 0\in P_ 0\), then \(\nu_ n\) converges weakly to \(R_ 0\). The method combines absolutely continuous changes of measures with a central limit theorem for the empirical distribution. As a by-product the authors obtain a central limit theorem for diffusion processes with mean field interaction.




Cited in
(36)








This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3341648)