Equivalence of ensembles, condensation and glassy dynamics in the Bose-Hubbard Hamiltonian (Q2283150)

The authors consider the Bose-Hubbard Hamiltonian without hopping terms, whose interactions are given by a classical non-interacting repulsive on-site term quadratic in the number of particles. Fixing the particle number density \(\rho\) and the energy density \(\epsilon\), they define a micro-canonical ensemble as the uniform measure on the set of classical distributions with the specified particle and energy density. If \(\epsilon\) is smaller than the ground state energy \(\epsilon_{\text{gs}}(\rho)\), which roughly scales as \(\rho^2\), then obviously no such state can exist. They show that there exists another critical line, given by \(\epsilon_{\text{c}}(\rho) = 2\rho^2+\rho \), such that, if \(\epsilon_{\text{gs}}(\rho) < \epsilon < \epsilon_{\text{c}}(\rho)\), then in the thermodynamic limit the micro-canonical ensemble converges to a unique Gibbs distribution of finite temperature (a grand-canonical ensemble with a given inverse temperature \(\beta>0\) and a fixed chemical potential \(\mu\)), while if \(\epsilon\) is above the critical line the micro-canonical ensemble converges to a unique infinite-temperature Gibbs state. Here convergence is in the sense that the relative entropy between the two distribution has a vanishing density as the volume goes to infinity, which implies convergence of expectation values of local observables. In the regime of high energy density, they furthermore show that the energy condensates in real-space, in the sense that for such micro-canonical ensembles, there is a site where a macroscopic number of particles is found in the limit, while the density of particles in the other sites vanishes. Finally, they argue in a non-rigorous way that such phase diagram should be preserved when the hopping terms is turned on. The reasoning is based on the fact that, under the grand-canonical ensemble, the hopping terms have zero expectation value, and thus the relationship between energy density and particle density should be independent of the hopping parameter. When \(\epsilon> \epsilon_{\text{c}}(\rho)\), under the assumption that the eigenstate thermalization hypothesis holds, the authors derive that eigenstates of the interacting Hamiltonian are superpositions of classical states with the corresponding energy density. Since the latter are shown by the rigorous results to condensate, the former should too. The authors observe that natural thermalization dynamics for the system in the consensation regime are exponentially slow, a phenomenon typical of glassy dynamics. While they do not rule out that faster processes could prepare such states, they present evidence that the presence of a condensate in the system should imply slow relaxation to the thermal equilibrium.