mathematics//game theory

2026-01-30

Game theory is the branch of mathematics that studies decisions among several players whose payoffs each depend on what all of them choose, and it is the frame an engineer needs when the other side adapts: an intruder probing a perimeter, a jammer facing a radio link, a competitor bidding in the same auction. The wind does not know a drone exists; an adversary does, and changes when the drone changes. Modelling it as noise assumes it never learns.

A game is written as players, the strategies each may choose, and the payoff each receives for every combination. In decision theory the uncertainty is a neutral nature with fixed probabilities and the rule is to minimize expected cost; once the other side decides too, and against you, the probabilities are no longer fixed, the expected value gives way to the worst case, and being predictable becomes a cost in itself.

Unpredictability pays when you are watched.

A guard who covers the more valuable of two gates with probability 2/3 cuts what an intruder can expect to take from 1 to 2/3, and announcing that randomized plan costs the guard nothing (mixed strategy).

When one side's gain is exactly the other's loss, the game is zero-sum, and the minimax theorem says that with randomized strategies the best guarantee a player can secure is the same whether the strategy is announced or kept secret. Robust design lives here: H-infinity control is a zero-sum game against a nature that picks the worst disturbance, and an adversarial example is the move of someone who has studied your network.

Outside zero-sum, the solution concept is the Nash equilibrium, a profile of strategies in which no player gains by changing only their own. It describes markets, auctions and shared channels, where interests partly align.

When one player commits first and the other responds having seen it, the game is a Stackelberg game, the situation of anyone who guards: Stackelberg models have been used to randomize patrols at US airports and ports, a real if niche deployment.

Repeated games play the same game many times, so reputation and retaliation enter; differential games play it in continuous dynamics, as in pursuit-evasion between an interceptor and a target (Isaacs, 1965). In defence these models are studied at textbook level, and multi-agent reinforcement learning against opponents that learn is research.