New math solves cooperation puzzle: exact formula for diverse groups

A team of researchers from Cardiff University and the University of Birmingham published an exact closed-form solution for cooperation probability in the heterogeneous public goods game. The paper, posted on arXiv in May 2026, tackles a classic problem in game theory: how cooperative behavior emerges when individuals differ in resources, productivity, and how quickly they adapt. The authors prove that under a condition called state-independence — where each player's payoff difference for switching strategy doesn't depend on others' current actions — the introspection Markov chain decomposes into N independent two-state chains, one per player. This factorization allows the stationary distribution to be expressed as a product measure, leading to the first exact formula for long-run cooperation probability without any asymptotic approximation.

The key result is a compact equation: p_C = (1/N) Σ_i [ (1 - μ_i0 - μ_i1) / (1 + exp(β_i α_i (1 - r_i/N))) + μ_i0 ], where α_i is each player's contribution, r_i the public goods multiplier, β_i selection intensity, and μ_i0/μ_i1 mutation rates. This yields several structural insights: a player-specific cooperation threshold at r_i = N (under symmetric mutation), meaning players only cooperate if their personal multiplier exceeds group size; payoff neutrality when selection intensity is zero; and clear sensitivity signs to each parameter. The result has implications for understanding collective action in biological systems, social networks, and multi-agent AI, where diverse agents must learn to cooperate. It also provides a benchmark for evaluating approximation methods and reinforcement learning algorithms in complex strategic settings.