Glauber dynamics for the quantum Ising model in a transverse field on a regular tree (Q2429395): Difference between revisions

Let \(G=(V,E)\) be a finite graph and \[ H = - \sum_{e =\{i,j\} \in E} \sigma_i^z\sigma_j^z - h\sum_{i\in V} \sigma_i^z- \lambda\sum_{i\in V} \sigma_i^x \] the Hamiltonian of the quantum Ising model at longitudinal and transversal external fields \(h\) and \(\lambda\geq 0\). The partition function and Gibbs state at inverse temperature \(\beta\) can be formulated in terms of a measure \(\mu_{G;\beta,h,\lambda}\) on \(\Sigma_{[0,\beta]}^V\) consisting of paths from \([0,\beta]\) to \(\{-1,1\}^V\), which is the reversible measure of a Glauber-type Markov process with state space \(\Sigma_{[0,\beta]}^V\). The dynamics consist of successive resamplings of whole spin-paths \((\sigma_i(t))_{t\in [0,\beta]}\). The article studies the measure and the Markov process. A first group of results, valid for general graphs, is about the monotonicity in \(h\) of the dynamics and equilibrium measure, the finiteness of the spectral gap and mixing time, and a generalization of a``censoring inequality'' of Peres and Winkler. A second group of results is about regular \(b\)-ary trees: the correlation length is bounded by \(\log b\), the Markov process for trees of depth \(\ell\) has a spectral gap bounded away from \(0\) and a mixing time growing at most linearly in \(\ell\). As a key tool, the authors analyze a resampling operator and cavity equation, dealing with additional difficulties present in the spin-paths setting and absent from classical spin lattices.