TB: Local vs Global Optima

A local optimum is the best solution in the neighborhood of the current position; a global optimum is the best solution across the entire solution space. Optimizing locally can trap a system at a local optimum, preventing it from reaching a global optimum that requires moving through a worse state first.

Local optimization is everywhere: evolution gets stuck in local optima (vestigial organs, suboptimal solutions that were good enough); companies optimize quarterly earnings at the expense of long-run value; individuals optimize for comfort at the expense of growth. The trap is invisible from inside a local optimum — everything looks like it's improving right up until you plateau. Escaping requires a willingness to get temporarily worse: take a step backward (unlearn, restructure, accept short-term losses) to find the path up the higher hill. This is why "disruption" works against incumbents who are stuck optimizing a local peak.

• No reading examples yet — will populate as books are ingested.

• Second-Order Thinking — escaping local optima requires thinking past the immediate next step
• Inversion — "what would keep me stuck at a local optimum?" surfaces the traps
• Bias-Variance Tradeoff — in ML, overfitting is the analogous trap: the model finds a local optimum on training data