Local vs Global Optima

Optimization finds local peaks—not necessarily the highest. The global optimum may be separated from your current position by valleys you'd have to descend. Hill-climbing strategies get stuck. This fundamental insight explains much apparent irrationality.

Imagine a fitness landscape: you want to reach the highest peak. You can only see your immediate surroundings. The obvious strategy: climb uphill. Always move toward higher ground.

This works—until you reach a local peak. From there, all directions are downhill. You stop. But you may not be on the highest peak. A higher mountain might be across a valley.

The Core Problem

Hill-climbing optimization:

• Evaluates local options
• Picks the best local option
• Repeats until no improvement possible

This converges to a local optimum—the best point from which no single-step improvement exists. But local optima can be far from global optima.

Getting to the global optimum might require temporarily going downhill—accepting worse outcomes to eventually reach better ones.

Where This Appears

Career

You've optimized for your current role. Lateral moves or temporary pay cuts feel like going downhill. But they might lead to higher peaks. Career local optima are common—people stuck in maximized positions that aren't globally best.

Business

"We can't cannibalize our existing product." Companies optimize current offerings incrementally. Disruption would require descending first. They get stuck; disruptors start in new valleys.

Evolution

Evolution is blind hill-climbing. It can't anticipate future fitness landscapes or coordinate multi-step moves. Organisms get stuck on local optima. Better designs exist but aren't reachable via single-step mutations.

Habits

You've optimized your daily routine for current goals. Changing requires going through an awkward transition period (valley). Many people stay on local optimum habits rather than endure the valley.

Relationships

A relationship can be locally optimal—best available given current network—but not globally optimal. Finding something better might require a period of searching (valley).

Why We Get Stuck

Local Visibility

We can't see the whole landscape. We don't know if our peak is highest. Without the big picture, we assume our local optimum is global.

Loss Aversion

Going downhill means accepting losses. Loss aversion makes the valley feel larger than it is. Even temporary losses trigger disproportionate resistance.

Uncertainty

The higher peak might not exist. Going downhill for nothing is worse than staying put. Uncertainty makes exploration risky.

Time Horizons

The valley is now; the higher peak is later. Short time horizons discount future gains. Immediate pain outweighs distant benefits.

Escaping Local Optima

Random Exploration

Occasionally take random steps instead of optimized ones. You might land in a better basin. (Simulated annealing in optimization theory.)

Restart From Different Points

Don't just optimize from where you are. Consider starting fresh from different positions. Multiple starting points explore more of the landscape.

Accept Temporary Losses

Budget for valleys. Expect that reaching higher peaks requires going through lower points. Plan resources for the descent.

Expand Visibility

Try to see more of the landscape. Talk to people on other peaks. Read broadly. Reduce the locality of your view.

Question "Optimization"

If a system is optimizing and stuck, ask: is this a local or global optimum? What valley would we need to cross? Is the crossing worth it?

— Decoded by DECODER