The voter model on modular networks asks how quickly groups with different initial opinions converge to consensus. The intuitive expectation is that stronger inter-group coupling accelerates agreement — more bridges between communities should speed convergence.
Note (2026-11-14; read against the paper's full text): the essay says that stronger coupling beyond an intermediate value slows consensus and that with a much smaller community the optimal coupling shifts lower. In the paper the intermediate optimum appears only for initially polarized and asymmetric populations; in the symmetric case the consensus time rises with alpha, the probability of a within-clique pairing, so more cross-clique interaction helps, and for most initial conditions the well-mixed case is the fastest route and modularity adds an alignment step without speeding the slow stage. The paper describes two stages: a fast alignment of the two cliques' opinion fractions, which alpha mainly affects, and a slow diffusion that is largely insensitive to alpha unless the cliques are nearly disconnected. When the minority shrinks (and under polarization) it reports that the optimum shifts toward smaller alpha, that is toward a larger share of cross-clique interactions and so stronger coupling, to speed the drift at the cost of larger fluctuations in the small clique; the essay's direction is the opposite.
The opposite occurs at an intermediate coupling strength. The consensus time is minimized not at maximum coupling and not at isolation, but at a specific intermediate value that depends on the asymmetry between groups. Stronger coupling beyond this point slows consensus rather than accelerating it.
The mechanism involves competing effects. Increasing cross-group coupling accelerates the drift toward agreement between groups, but it simultaneously amplifies stochastic fluctuations within the smaller group by coupling it to the larger group's internal noise. Below the optimal coupling, drift dominates and adding bridges helps. Above it, noise amplification dominates and additional bridges hurt. The minimum occurs where these two effects are balanced.
Group size asymmetry sharpens the effect. When communities are equal, the optimal coupling is relatively insensitive. When one community is much smaller, the noise amplification effect is stronger — the small group is overwhelmed by the larger group's fluctuations — and the optimal coupling shifts lower. The most polarized configurations are the most sensitive to over-coupling.
The structural observation: a quantity assumed to be monotonic (consensus speed as a function of coupling) has an interior optimum. The bridge that connects communities serves two functions simultaneously — transmitting signal and transmitting noise — and the noise channel grows faster than the signal channel beyond a threshold. The optimal bridge is not the strongest one.