Capturing long range correlations in two-dimensional quantum lattice systems using correlator product states

We study the suitability of correlator product states for describing ground-state properties of two-dimensional spin models. Our ansatz for the many-body wave function takes the form of either plaquette or bond correlator product states and the energy is optimized by varying the correlators using Monte Carlo minimization. For the Ising model we find that plaquette correlators are best for estimating the energy while bond correlators capture the expected long-range correlations and critical behavior of the system more faithfully. For the antiferromagnetic Heisenberg model, however, plaquettes outperform bond correlators at describing both local and long-range correlations because of the substantially larger number of local parameters they contain. These observations have quantitative implications for the application of correlator product states to other more complex systems, and give important heuristic insights: in particular the necessity of carefully tailoring the choice of correlators to the system considered, its interactions and symmetries.