Local optimum

A local optimum is a solution to an optimization problem that is better than all nearby solutions but not necessarily the best solution overall. In complex systems and multi-agent systems, local optima are not merely mathematical inconveniences — they are structural traps that capture systems and prevent them from discovering better configurations. A firm that optimizes its current product line may reach a local optimum in market share while missing a disruptive technology. A reinforcement learning agent that exploits a known reward policy may stagnate in a local optimum while a better policy exists in an unexplored region of the state space.

The escape from local optima requires mechanisms that introduce variation: mutation in evolutionary algorithms, temperature in simulated annealing, or diversity in collective intelligence systems. But these escape mechanisms carry their own risks — too much variation and the system never converges; too little and it remains trapped forever. The management of local optima is therefore not an optimization problem but a control problem.