In a one-dimensional exclusion process with look-ahead rules, a particle can jump forward I sites if all intermediate sites are empty. This is a microscopic model for traffic: a car can accelerate into the gap ahead only if there's nothing in the way. The hopping rates depend on local spacing, following Arrhenius dynamics.
The conventional requirement for a stationary distribution is detailed balance — every transition and its reverse occur at equal rates. Lam, Ngo, and Huynh show that these look-ahead processes satisfy something weaker: pairwise balance. Pairs of configurations reach equilibrium without the full global balance condition holding. The result: the invariant measure takes the Ising-Gibbs form, but the stationarity mechanism is fundamentally non-equilibrium.
The stationary current in the thermodynamic limit recovers the mean-field prediction for traffic flow exactly when particles are uncorrelated. When interactions are nontrivial, the framework provides the exact correlation-induced correction. The mean-field approximation isn't wrong — it's the zero-correlation limit of a more general result.
The structural insight: balance doesn't have to be global to produce a steady state. Pairwise balance — each pair of configurations in mutual equilibrium — is sufficient for the whole system to reach stationarity. And once you have the measure, you can compute exactly how correlations modify the macroscopic current. The mean-field theory is the starting point, not the approximation. Correlations are the correction, and the correction is exact.