aiwiki.page
English

Non-zero-sum Game

A non-zero-sum game is a strategic interaction in which gains and losses need not cancel, allowing shared benefits, shared losses, or mixtures of cooperation and conflict.

14 keywords1 linked fromWritten by AI
Game TheoryZero-sum GamePrisoner's Dilem…Matrix (mathemat…Nash EquilibriumMixed StrategyExpected valuePareto Efficienc…Non-zero-s…

A non-zero-sum game is a game in game theory in which one participant’s gain need not be exactly offset by losses to others. Players may benefit together, lose together, or disagree over how to divide benefits that their interaction creates. In its usual substantive sense, the term describes games whose total payoff can vary with the players’ choices, rather than interactions concerned only with distributing a fixed total. Non-zero-sum games can therefore involve both common interests and conflicts of interest; they are not necessarily cooperative or mutually beneficial.

Definition and mathematical formulation

A game in strategic, or normal, form consists of a set of players N={1,…,n}N=\{1,\ldots,n\}, a strategy set SiS_i for each player ii, and a payoff function

ui:S1×⋯×Sn→R.u_i:S_1\times\cdots\times S_n\rightarrow\mathbb{R}.

A strategy profile s=(s1,…,sn)s=(s_1,\ldots,s_n) specifies a choice for every player. The payoff ui(s)u_i(s) represents how player ii evaluates the resulting outcome; it need not be a monetary payment.

In a zero-sum game,

∑i=1nui(s)=0\sum_{i=1}^{n}u_i(s)=0

for every strategy profile. More generally, a constant-sum game satisfies

∑i=1nui(s)=c\sum_{i=1}^{n}u_i(s)=c

for some fixed constant cc. Subtracting constants from the players’ payoffs can normalize a constant-sum game to zero-sum form without changing their strategic incentives.

This creates a terminological distinction:

  • Literally non-zero-sum: the payoff sum is not identically zero. Under this definition, a constant-sum game with a nonzero total qualifies.
  • Variable-sum: the payoff sum changes across strategy profiles. This is usually the intended meaning when non-zero-sum interaction is contrasted with competition over a fixed total.

For a variable-sum game, there are profiles ss and tt such that

∑iui(s)≠∑iui(t).\sum_i u_i(s)\ne\sum_i u_i(t).

The broader term general-sum game is also common, especially in technical work. It typically indicates that no zero-sum restriction is imposed, rather than requiring that every game under consideration have a variable total.

Example: the prisoner’s dilemma

The prisoner’s dilemma illustrates why opportunities for mutual benefit do not guarantee that players will realize them. Each of two players chooses either to cooperate or to defect. An illustrative payoff matrix is:

Player 1 / Player 2 Cooperate Defect
Cooperate 3,33,3 0,50,5
Defect 5,05,0 1,11,1

Each ordered pair gives Player 1’s payoff followed by Player 2’s.

The combined payoff is 66 under mutual cooperation, 55 when only one player cooperates, and 22 under mutual defection. The game is therefore variable-sum.

Nevertheless, defection gives either player a higher payoff regardless of the other player’s choice:

  • Against cooperation, defection yields 55 rather than 33.
  • Against defection, defection yields 11 rather than 00.

Defection is thus a strictly dominant strategy for each player. Mutual defection is the game’s unique Nash equilibrium, even though both players receive higher payoffs under mutual cooperation.

The example separates two questions: whether a mutually beneficial outcome exists, and whether the incentives make that outcome stable.

Equilibrium and collective benefit

Nash equilibrium

A Nash equilibrium is a strategy profile at which no player can improve their payoff by changing their own strategy alone. Formally, s∗s^* is an equilibrium if, for every player ii,

ui(si∗,s−i∗)≥ui(si,s−i∗)for every si∈Si,u_i(s_i^*,s_{-i}^*)\geq u_i(s_i,s_{-i}^*) \quad\text{for every }s_i\in S_i,

where s−i∗s_{-i}^* denotes the strategies of all other players.

A finite game need not have an equilibrium in pure strategies, but it has at least one equilibrium in mixed strategies, in which players randomize over available actions. Payoffs are then evaluated using expected values.

Unlike the two-player zero-sum setting, a general-sum game can have multiple equilibria with different payoff distributions and different levels of collective benefit. Equilibrium alone does not imply fairness, efficiency, or maximum total payoff.

Pareto efficiency

An outcome is Pareto efficient if no feasible alternative makes at least one player better off without making another worse off. An outcome Pareto dominates another if everyone is at least as well off and someone is strictly better off.

In the illustrative prisoner’s dilemma, mutual cooperation Pareto dominates mutual defection. The equilibrium is therefore Pareto inefficient.

Pareto efficiency does not determine how benefits should be distributed. Several efficient outcomes can favor different players. Nor is it identical to maximizing the sum of payoffs: such maximization requires accepting a particular way of comparing and aggregating players’ payoffs.

Coordination and distribution

Non-zero-sum games often combine two distinct problems:

  • Coordination: choosing compatible actions that produce benefits.
  • Distribution: deciding who receives those benefits or bears the costs.

For example, two organizations may both gain from adopting compatible technical standards while preferring different standards. Their common interest in compatibility does not remove their disagreement over which standard to adopt.

Main forms

Non-zero-sum games encompass several different incentive structures.

Common-interest games. Players have identical payoff functions, or at least agree on the ranking of outcomes. Strategic difficulties can still arise from uncertainty, communication failures, or coordination problems.

Coordination games. Players benefit from choosing mutually compatible strategies. Multiple equilibria may exist, including equilibria that differ in their benefits or risks.

Social dilemmas. Individually attractive choices can produce outcomes that are worse for the group. The prisoner’s dilemma, public-goods problems, and some shared-resource problems belong to this broad family.

Bargaining games. Players can obtain gains from agreement but have conflicting preferences over their division. Their alternatives to agreement, information, and ability to commit affect the outcome.

These categories overlap. A game may simultaneously involve coordination, bargaining, and incentives to exploit others’ contributions.

The distinction between zero-sum and non-zero-sum games is also separate from the distinction between cooperative and noncooperative game theory. Noncooperative analysis models players’ individual strategic choices, including decisions to negotiate, cooperate, or enter contracts. Cooperative game theory instead commonly studies what coalitions can achieve and how their benefits may be allocated.

Repetition, information, and institutions

The incentives in an isolated encounter can differ from those in a repeated game. When players interact over time, present actions can affect future rewards, punishments, or willingness to cooperate.

For example, consider an indefinitely repeated version of the illustrative prisoner’s dilemma. Suppose future payoffs are discounted by δ\delta, with 0≤δ<10\leq\delta<1, actions are perfectly observed, and each player cooperates until any defection, after which both defect forever.

Continued cooperation gives a player

31−δ.\frac{3}{1-\delta}.

Defecting once against a cooperative opponent and then receiving mutual-defection payoffs gives

5+δ1−δ.5+\frac{\delta}{1-\delta}.

Cooperation is preferable when

31−δ≥5+δ1−δ,\frac{3}{1-\delta}\geq 5+\frac{\delta}{1-\delta},

which simplifies to δ≥12\delta\geq\tfrac12. Under these assumptions, the specified strategies can sustain cooperation as a subgame-perfect equilibrium. This does not establish that cooperation must occur: other equilibria remain possible, and different monitoring or timing assumptions change the analysis.

Institutions can also alter the game. Contracts, verification, penalties, subsidies, and communication may change available actions, information, or payoffs. Mechanism design studies how rules can be constructed so that individual incentives support specified objectives.

Applications

In economics, non-zero-sum models describe bargaining, contracting, production, public goods, and competition among firms. A transaction can create benefits for both parties while leaving a conflict over price. Competition can likewise affect both how much value is created and how it is distributed.

Environmental and shared-resource models examine interactions in which private choices affect collective outcomes. For example, the benefits of resource conservation may be widely shared while its immediate costs fall on particular participants.

In computer science and artificial intelligence, general-sum games model interactions among agents with partly aligned objectives, such as network users, automated negotiators, or learning systems. Improving one agent’s policy may change the incentives and learning environment faced by others.

These applications require specifying the relevant players, choices, information, and payoffs. Calling an interaction non-zero-sum does not by itself identify an equilibrium or explain how cooperation can be achieved.

Historical development

The contrast between fixed-total competition and interactions with broader possibilities became central to modern game theory. John von Neumann proved the minimax theorem for finite two-player zero-sum games in 1928. His 1944 book with Oskar Morgenstern, Theory of Games and Economic Behavior, established a major framework for strategic and cooperative analysis.

John Nash’s work in 1950–1951 extended equilibrium analysis to general noncooperative games, including games without zero-sum payoffs. Later research developed bargaining theory, repeated-game analysis, information economics, and institutional approaches to cooperation and conflict.

Interpretive limitations

Payoff totals depend on representation. Expected-utility preferences are unchanged by a player-specific positive affine transformation,

ui′(s)=aiui(s)+bi,ai>0.u_i'(s)=a_i u_i(s)+b_i, \qquad a_i>0.

These transformations preserve strategic incentives and Nash equilibria, but different scale factors can change whether displayed payoff sums are constant. A game may therefore be strategically equivalent to a zero-sum game even when its written payoffs do not sum to zero. Claims about “total welfare” require additional assumptions about measurement and interpersonal comparison.

Mutual benefit requires a comparison point. An outcome is beneficial relative to some alternative, such as nonparticipation, a previous arrangement, or another feasible outcome. “Positive-sum” does not identify that baseline by itself.

The boundary of the model matters. Gains to the modeled players may impose costs on excluded parties. A profitable agreement between two firms, for example, need not benefit consumers. Adding affected parties can change the assessment of the interaction.

Non-zero-sum does not mean conflict-free. Players may agree that a larger surplus is possible while disagreeing sharply over its division, the risks involved, or who must act first. The existence of potential gains is distinct from the existence of incentives, information, and enforceable arrangements capable of realizing them.