Game theory is the mathematical study of strategic decision-making among rational agents. From the Prisoner's Dilemma to evolutionary biology, Nash equilibria reveal why individually rational choices can produce collectively disastrous outcomes — and how cooperation can nonetheless emerge.
A game in the mathematical sense is a model of any situation where multiple agents make decisions and each agent's outcome depends not only on their own choice but also on the choices of others. Three primitives define every game: the set of players, the set of available strategies for each player, and a payoff function that maps every combination of strategy choices to an outcome value for each player.
For two-player games, the payoff structure is conveniently represented as a payoff matrix: rows correspond to Player 1's strategies, columns to Player 2's strategies, and each cell contains the ordered pair (Player 1 payoff, Player 2 payoff) that results from that combination. Reading the matrix allows us to identify which strategies are attractive for each player given their beliefs about what the other will do.
A dominant strategy is one that yields a higher payoff for a player regardless of what any other player does. When a dominant strategy exists, a rational player will always choose it — no prediction about opponents is necessary. Not all games have dominant strategies, but when they do, the analysis simplifies dramatically.
The most famous game in all of game theory is the Prisoner's Dilemma. Two suspects are interrogated separately. Each can either Cooperate (stay silent, denoted C) or Defect (betray the other, denoted D). The payoffs in years of prison sentence avoided are:
If both stay silent, each gets one year (cooperative outcome, payoff -1 each). If one betrays while the other stays silent, the betrayer goes free (payoff 0) while the loyal partner gets three years (-3). If both betray, both get two years (-2 each). Now consider Player 1's reasoning: if Player 2 cooperates, Player 1 gets -1 by cooperating but 0 by defecting — defecting is better. If Player 2 defects, Player 1 gets -3 by cooperating but -2 by defecting — defecting is still better. Defection is a dominant strategy for both players. Yet the outcome (D, D) with payoffs (-2, -2) is worse for everyone than mutual cooperation (-1, -1). This is the core tension in all social dilemmas: individual rationality leads to collective inefficiency.
The central solution concept of non-cooperative game theory is the Nash equilibrium, named for mathematician John Nash who proved its existence in 1950. Formally, a strategy profile (s1*, s2*, ..., sn*) is a Nash equilibrium if no single player can increase their own payoff by unilaterally switching to a different strategy, holding all other players' strategies fixed. Each player's strategy is a best response to the strategies of all other players.
Nash's existence theorem guarantees that every finite game — one with a finite number of players and finite strategy sets — has at least one Nash equilibrium, possibly in mixed strategies where players randomize over their pure strategies. This result, proved using Kakutani's fixed-point theorem, is one of the foundational achievements of 20th-century mathematics.
In the Prisoner's Dilemma, (Defect, Defect) is the unique Nash equilibrium. If Player 1 defects, Player 2's best response is also to defect (since -2 > -3). Symmetrically for Player 2. Neither player can improve their payoff by switching unilaterally — the equilibrium is stable, even though it is Pareto-inferior to (Cooperate, Cooperate).
Mixed strategy Nash equilibria arise when no pure strategy is always best. In Rock-Paper-Scissors, any pure strategy can be beaten, so the unique Nash equilibrium has each player randomizing uniformly: each action played with probability 1/3. The logic is that a player will mix only if they are indifferent between the pure options — and that indifference is precisely maintained by the opponent's mixing probabilities.
When multiple equilibria exist, game theory faces the equilibrium selection problem. In the Stag Hunt, two hunters can each choose to hunt a stag together (high reward, requires coordination) or hunt a rabbit alone (guaranteed but lower reward). Both (Stag, Stag) and (Rabbit, Rabbit) are Nash equilibria. The former is payoff-dominant; the latter is risk-dominant. Which equilibrium players reach depends on beliefs, communication and focal points — questions game theory alone cannot fully resolve.
The one-shot Prisoner's Dilemma traps players in defection. But what happens when the same players meet repeatedly? The iterated Prisoner's Dilemma changes everything: reputation, reciprocity and the shadow of the future all become strategically relevant. A player who defects today risks retaliation tomorrow, making cooperation potentially sustainable as a rational long-run strategy.
Political scientist Robert Axelrod ran a famous series of computer tournaments in 1980, inviting game theorists to submit strategies for the iterated Prisoner's Dilemma. The simplest strategy submitted — Tit-for-Tat — won both tournaments. Tit-for-Tat cooperates on the first move and then mirrors whatever the opponent did last round. It is nice (never defects first), retaliatory (punishes defection immediately), forgiving (returns to cooperation as soon as the opponent does), and clear (easy for opponents to learn). These four properties explain its robustness across a wide range of competing strategies.
Evolutionary game theory, pioneered by John Maynard Smith and George Price, asks not which strategy is individually rational, but which strategies can survive in a population subject to natural selection. A strategy is an Evolutionarily Stable Strategy (ESS) if a population of players using it cannot be invaded by a small group of mutants using any alternative strategy. Formally, a strategy I is an ESS if for any mutant strategy J:
The Hawks and Doves model illustrates this beautifully. Hawks always escalate conflicts over resources and fight until one party is injured. Doves always display but flee if the opponent escalates. In a population of all Doves, a Hawk mutant does very well (it wins every contest without injury). In a population of all Hawks, a Dove mutant does better than average (it avoids costly fights). Neither pure strategy is an ESS; instead, a stable mixed-frequency equilibrium emerges in which both types coexist. The ESS frequency of Hawks is p* = V/C when V < C, where V is the value of the resource and C is the cost of losing a fight.
Crucially, cooperation can also evolve in spatial settings even without kin selection or repeated interaction. When agents play on a lattice and interact only with neighbours, cooperators can form clusters that shield each other from exploitation. Local interaction geometry enables cooperation to persist where it would be swept away in a well-mixed population.
Game theory has become indispensable across virtually every social and natural science. Its mathematical framework converts qualitative intuitions about strategic behaviour into precise, testable predictions.
In industrial organization, the Bertrand and Cournot models use Nash equilibrium to predict prices and quantities in oligopolistic markets. Auction design relies heavily on game theory: the Vickrey (second-price) auction has the remarkable property that bidding your true value is a dominant strategy, making truth-telling individually rational. The Nobel Prize-winning field of mechanism design (Myerson, Maskin, Hurwicz, 2007) asks the inverse question: given a desired social outcome, what game rules will make self-interested agents produce it?
Voting systems exhibit strategic manipulation: the Condorcet paradox shows that majority preference can cycle (A beats B, B beats C, C beats A) with no stable winner. Nuclear deterrence during the Cold War is analyzed as a Nash equilibrium: Mutual Assured Destruction (MAD) is stable precisely because neither side can gain by launching first — a grim game-theoretic argument for stability through catastrophic threat. Coalition formation in parliaments uses cooperative game theory and the Shapley value to assign power to parties.
Animal contests, honest signalling (the Zahavian handicap principle — costly signals are reliable because they cannot be faked cheaply), and the co-evolution of parasites and hosts are all modelled as evolutionary games. Sexual selection, altruism among kin (Hamilton's rule: rb > c), and the evolution of spite all have game-theoretic explanations grounded in fitness payoffs.
Braess's paradox is one of the most counterintuitive results in algorithmic game theory: adding a new road to a network can slow down all drivers when each routes selfishly. The reason is that the new road shifts Nash equilibrium traffic flows in ways that increase total congestion. Internet search and display advertising use Vickrey-Clarke-Groves (VCG) mechanisms to incentivize truthful bidding in real-time auctions run billions of times per day.
Browse interactive algorithm and game-theory simulations — see agents cooperate, compete and evolve strategies in real time.
Game theory splits into non-cooperative game theory (analyzing strategic interaction between self-interested players without binding agreements — the basis for most economics) and cooperative game theory (studying coalition formation and fair distribution when binding agreements are possible). Other branches include evolutionary game theory (applying game theory to biological evolution), mechanism design (reverse game theory — designing rules to achieve desired outcomes), and repeated games (analyzing long-term strategic interaction).
A zero-sum game is one where the total payoff to all players is constant — one player's gain exactly equals another's loss. Chess, poker, and competitive markets for fixed-supply goods are approximately zero-sum. Minimax theorem (von Neumann 1928) guarantees every two-player zero-sum game has an optimal mixed strategy solution. Most real economic and social situations are non-zero-sum — cooperation can create value for all parties.
Backward induction is a solution method for sequential games (extensive form). Starting from the final decision node, determine the optimal action. Then work backward to the previous node, assuming future play will be optimal. Continue to the root. This procedure finds the subgame perfect Nash equilibrium — a set of strategies that are Nash equilibria in every subgame, eliminating non-credible threats. It solves chess in principle (though impractical due to game tree size).
The Folk Theorem states that in infinitely repeated games, any individually rational outcome (where each player gets at least their minimax payoff) can be sustained as a Nash equilibrium, given players are sufficiently patient (discount factor close to 1). This explains how long-term relationships support cooperation that would be impossible in one-shot games: the threat of future punishment sustains current good behavior.
Auction theory applies game theory to auctions — mechanisms for allocating goods to bidders. Key auction formats include English (ascending price), Dutch (descending price), first-price sealed-bid, and Vickrey (second-price sealed-bid). The Revenue Equivalence Theorem shows these generate equal expected revenue under symmetric conditions. Vickrey auctions have dominant truth-telling strategies. Auction theory is applied to spectrum rights, online advertising (Google AdWords), and public procurement.
In perfect information games (chess, checkers), all players know the complete history of the game and the entire game tree. In imperfect information games (poker, Diplomacy), players have private information — their own cards — that others don't know. Imperfect information introduces strategic bluffing and signaling. Game theory handles imperfect information using Bayesian Nash equilibrium (players optimize over beliefs about others' private information).
A mixed strategy Nash equilibrium involves players randomizing over their pure strategies according to probability distributions. Even games with no pure-strategy Nash equilibrium always have a mixed-strategy equilibrium (Nash's theorem). Players mix when each pure strategy they use yields equal expected payoff. In tennis, the optimal serve strategy involves mixing between body serves and wide serves so opponents can't exploit predictability.
The Shapley value is a cooperative game theory solution concept that fairly distributes the total value created by a coalition among players based on their marginal contributions. Each player receives their average marginal contribution across all possible coalition formation orders. The Shapley value satisfies four axioms: efficiency (distributes all value), symmetry (equal players get equal shares), linearity, and dummy (players who add no value get nothing). It's used in cost sharing, voting power analysis, and machine learning feature attribution (SHAP values).
The principal-agent problem (contract theory) studies situations where a principal (employer, shareholder) delegates work to an agent (employee, manager) with different information and possibly different interests. The agent may shirk (moral hazard) or misrepresent their type (adverse selection). Optimal contract design aligns incentives through monitoring, performance pay, options, and reputation mechanisms. Key applications: CEO compensation, insurance deductibles, franchise agreements.
Game theory models nuclear deterrence (Mutually Assured Destruction as a Nash equilibrium), trade negotiations (repeated Prisoner's Dilemma with reputation), arms races (coordination games), alliance formation (coalition theory), and sanctions regimes. Schelling's 1960 "The Strategy of Conflict" pioneered using game theory for international security. The Cuban Missile Crisis has been extensively analyzed as a game of chicken where both sides' resolve and communication shaped the equilibrium.