Convergence of Gibbs measures associated with simulated annealing: The case of distance squared (Q2371910)

Let \(P_\lambda(B)= {\int_B e^{-\lambda J(x)}dx\over \int_{\mathbb{R}^n} e^{-\lambda J(x)}dx}\), where \(J(x)> 0\) is a continuous function such that the denominator of \(P_\lambda(B)\) is finite, and \(B\) is a Borel set. Therefore \(P_\lambda(B)\) is a probability measure, called a Gibbs measure. For a sequence of probability mesures \(\{P_n\}\) if \(\int\phi dP_n\to \int\phi dP\) for all bounded continuous real-valued function \(\phi\), we say that \(\{P_n\}\) convergence weakly to a probability measure \(P\). The author states a theorem from the paper by Dennis Cox, Robert Hardt and Petr Kloucek as the following result: Assume 1. \(J\in C^3(\mathbb{R}^n)\), 2. \(J\geq 0\), 3. \(J(x)\geq\| x\|^p\) for \(x\) sufficiently large, \(p> 0\), 4. \(M= \{x\in \mathbb{R}^n: J(x)= 0\}\) is nonempty and bounded, 5. The Hessian \(D^2J\) has constant rank \(k\) near \(M\). The \(M\) is a \(C^{2,x}\) \(n-k\)-dimensional manifold, and \(P_\lambda\to P\) as \(\lambda\to \infty\).