Mean-field approximation and phase transitions in an Ising-voter model on directed regular random graphs

It is known that on directed graphs, the correlations between neighbors of a given site vanish and thus simple mean-field-like arguments can be used to describe exactly the behavior of Ising-like systems. We analyze heterogeneous modifications of such models where a fraction of agents is driven by either voter or antivoter dynamics, and align (voter) or antialign (antivoter) with a randomly chosen out-neighbor. It turns out that voter agents do not affect the dynamics of the model, and it behaves like a pure Ising model. Antivoter agents have a stronger impact since they act as a kind of noise, which weakens a ferromagnetic ordering. Only when Ising spins are driven by the heat-bath dynamics, the behavior of the model is correctly described by the mean-field approximation. The Metropolis dynamics generates some additional correlations that render the mean-field approach approximate. Simulations on annealed networks agree with the mean-field approximation but for the model with antivoters and with the Metropolis dynamics only its heterogeneous version provides such an agreement. Calculation of the Binder cumulant confirms that critical points in our models with the heat-bath dynamics belong to the Ising mean-field universality class. For the Metropolis dynamics, the phase transition is most likely discontinuous, at least for not too many antivoters.