Statistical mechanics of a three-body Hopfield model with finite connectivity
A three-body Hopfield model defined on sparse random graphs with finite mean connectivity is studied using a replica-symmetric (RS) analysis. By extending the functional replica framework for conventional two-body finite-connectivity Hopfield models, self-consistent equations for the local-field distribution are derived and solved numerically by population dynamics. At low memory load, the RS analysis shows a discontinuous transition between paramagnetic and retrieval states, accompanied by coexistence of distinct fixed points and an upper spinodal of the retrieval solution. Starting the RS population dynamics from an uninformative finite-amplitude field distribution also leads to a spin-glass-like (SG-like) fixed point at low temperatures, even though the paramagnetic fixed point remains locally stable. As the number of embedded patterns increases, the retrieval solution disappears, while the SG-like RS solutions are found at high load. However, the local instability and thermodynamic inconsistencies of the low-temperature RS solutions indicate the limitations of the RS description of the glassy regime. Finite-size Monte Carlo simulations using population annealing agree well with the RS description of the retrieval branch during heating but exhibit glassy trapping behavior under cooling. Such trapping can occur even for a single embedded pattern, where conventional cross-talk noise among stored patterns is absent, indicating the importance of sparse connectivity and three-body interaction. The characteristic number of embedded patterns separating different solution regimes grows approximately linearly with the mean connectivity, corresponding to finite characteristic load ratios. These results clarify how sparse connectivity and many-body interactions influence the stability and accessibility of retrieval states in associative memory.