Role of the memory in convergence to invariant Gibbs measure (Q393845): Difference between revisions

Problems of the equilibrium states for Hamiltonian manyparticle systems under external random forces are very important for physics backgrounds and are strongly related with mixing invariant sbmanifolds giving rise to stationary invariant probabilistic merasures. If the external forces \( f_{t}\) are of the white noise type, these equilibrium measures coincide with the Gibbs one, having no time memory. Otherwise, the corresponding measures are typically different from the Gibbs measure and the work is devoted to this case. In particular, for a finite dimensional Hamiltonian system with \[ H_{N}(\psi ):=\frac{1}{2}\sum_{j=1}^{N}p_{j}^{2}+\frac{1}{2}V(i,j)q_{i}q_{j}, \] for which \(L_{-}^{\perp }=\{0\},\) where \[ L_{-}:=\{\psi \in \mathbb{E }^{N}\times \mathbb{E}^{N}:H_{N}(e^{tA}\psi )\rightarrow 0 \,\, \text{if}\, \, t\rightarrow \infty \}, \] then for almost all \(C_{f}:=<f_{t}f_{t+s}>\in S( \mathbb{R})\) there are nonzero correlations between velocities, \(C_{\psi }(p_{i},p_{j})\neq 0\) for some \(i\neq j\in \overline{1,N},\) and the limiting distribution is not Gibbs. Similar results are also obtained for the case of infinite-dimensional Hamiltonian systems when \(N\rightarrow \infty\), and the particles are infinitely far away from the boundary.