In World War I, frontline soldiers on opposing sides spontaneously established local truces — the “live-and-let-live” system. Units facing each other across no man's land would tacitly agree not to shoot to kill, firing at predictable times in predictable directions so the other side could avoid the shells. The cooperation was stable, widespread, and entirely unauthorized.
Command authorities responded by destroying it. Raids were ordered — small-scale assaults that forced engagement and broke the cooperative equilibrium. Officers who tolerated quiet sectors were rotated out. The punishment for not fighting was made more costly than the fighting itself.
This paper formalizes the mechanism using continuous-strategy evolutionary game theory. Local units can converge on cooperative equilibria through repeated interaction — the standard result from iterated prisoner's dilemmas. But the model adds a hierarchical layer: a central authority that imposes enforcement penalties against cooperation. The authority's incentives need not align with the soldiers' welfare; they may derive political returns from continued conflict.
The result: stable peace requires one of three conditions. Either the central authority's incentives must align with frontline welfare (the authority wants peace too), or external constraints must limit the authority's enforcement capacity (it cannot punish cooperation), or the political returns to conflict must diminish (continuing the war becomes more costly than ending it). Without at least one of these, the hierarchy will continuously dismantle any locally emergent cooperation.
The lesson extends beyond warfare. Any hierarchical system where local agents can cooperate but central authorities benefit from competition will exhibit the same pattern: cooperation forms from below and is destroyed from above.