Kakutani fixed-point theorem

Overview

The Kakutani fixed-point theorem guarantees the existence of a “fixed point” for set-valued mappings (correspondences) that are well behaved on compact, convex sets. A fixed point here is a point x such that x is contained in the set Φ(x) that the correspondence assigns to x. In economics, this result is the workhorse for proving the existence of equilibria when individual best responses may be non-unique—most notably, Nash equilibria (including mixed strategies) in games and competitive equilibria in general equilibrium models. It extends Brouwer’s fixed-point theorem from single-valued continuous functions to set-valued, upper hemicontinuous correspondences with convex values.

Origins and Credit

Shizuo Kakutani proved the theorem in 1941, generalizing Brouwer’s 1911 result. John Nash (1950–1951) famously used Kakutani to prove that every finite normal-form game possesses a mixed-strategy equilibrium. Arrow, Debreu, and McKenzie later relied on related fixed-point arguments in establishing competitive equilibrium existence. The theorem is now a staple in graduate microeconomics, game theory, and mathematical economics.

Core Idea and Mechanics

A correspondence Φ maps each point x in a set K to a set of points Φ(x) rather than a single point. The Kakutani theorem states, roughly:

If K is a nonempty, compact, convex subset of Euclidean space, and Φ: K ⇉ K is a correspondence such that for every x in K, Φ(x) is nonempty, convex, and compact; and Φ is upper hemicontinuous (intuitively, small changes in x do not cause the image sets to jump outward), then there exists x* in K with x* ∈ Φ(x*).

Two pieces of jargon in plain language:

• Upper hemicontinuity (UHC): As x varies slightly, the set Φ(x) cannot suddenly sprout distant new points; the possible images vary “tighteningly.” Formally, UHC is implied by having a closed graph (whenever x^k→x and y^k→y with y^k∈Φ(x^k), then y∈Φ(x)) and compact values.
• Convex-valued: Each image Φ(x) is convex; for any two feasible responses y and z, their mixtures ty + (1−t)z are also feasible. This convexity is what randomization or “mixing” strategies delivers in games.

Why economists care: best-response “functions” are often not functions. A firm may be indifferent among a range of outputs or prices that maximize profit; a player in a game may have multiple best responses. Stacking these best responses yields a set-valued map from choices into choices. Kakutani ensures a fixed point of this map—a profile where everyone’s choice is simultaneously a best response—i.e., an equilibrium.

Key Assumptions and Conditions

• Domain: A nonempty, compact, convex subset K of ℝⁿ (for example, a probability simplex of mixed strategies or a price simplex).
• Self-mapping: Φ maps K into itself: Φ(x) ⊆ K for all x ∈ K.
• Nonempty, compact, convex images: For each x, Φ(x) is nonempty (there is at least one best response), bounded/closed (compact), and convex (mixing is feasible).
• Upper hemicontinuity: The correspondence varies without outward jumps as x changes; having a closed graph with compact values is a common sufficient condition.

Implications

• Nash equilibrium existence: In finite games, allowing mixed strategies makes each player’s best-response correspondence nonempty, convex, and upper hemicontinuous; the joint strategy space is a compact simplex. Kakutani then delivers a mixed-strategy equilibrium (Nash’s theorem).
• General equilibrium: Demand and supply correspondences are set-valued when optima are non-unique. Under convex preferences/technologies and continuity, a normalized price simplex and an excess-demand or price-adjustment correspondence satisfy Kakutani’s conditions, yielding a competitive equilibrium.
• Modeling discipline: When best responses might be multi-valued (flat profit regions, kinks), ensuring convex feasible sets, compactness (bounds), and continuity allows existence to go through.
• Design levers: Introducing randomization, mixed strategies, or lottery over allocations can convexify choice sets, restoring existence in otherwise nonconvex environments.
• Foundation for robustness: Many existence results (e.g., Debreu–Fan–Glicksberg for continuous games) rely on Kakutani or its variants; once existence is assured, comparative statics can be studied with additional tools.

Example in Practice

Price competition with flat best responses. Two brands choose prices p₁, p₂ in [p_min, p_max]. Demand shares are continuous in prices, but due to a retailer policy, profits are piecewise linear with plateaus (for a range of prices, the same shelf placement yields the same sales). Each firm’s profit is continuous and quasi-concave in its own price, so the best-response correspondence BR_i(p_−i) is nonempty, compact, and convex (an interval) for any rival price. As rivals’ prices change slightly, these intervals move without sudden outward jumps—i.e., BR is upper hemicontinuous.

Define the joint best-response correspondence BR(p₁, p₂) = BR₁(p₂) × BR₂(p₁) mapping the compact rectangle [p_min, p_max]² into itself. All Kakutani conditions hold, so there exists (p₁*, p₂*) with (p₁*, p₂*) ∈ BR(p₁*, p₂*). That pair is a Nash equilibrium even though each firm may have had multiple best responses. For a pricing team, this assures that at least one internally consistent price pair exists given the plateau-like incentives created by retail contracts.

Limitations and Common Misunderstandings

• Existence, not uniqueness or stability: Kakutani guarantees at least one fixed point; there may be many. It says nothing about convergence of adjustment processes or algorithms.
• Not constructive: The theorem proves existence but does not provide a computational method. Additional structure (contractions, potential functions) is needed for algorithmic guarantees.
• Convexity and compactness are substantive: Discrete choices, fixed costs, or increasing returns can make best-response sets nonconvex or unbounded, breaking the conditions. Remedies include randomization, capacity bounds, or alternative equilibrium concepts.
• Continuity/UHC requirements: Discontinuities (e.g., hard thresholds, discontinuous payoffs) can violate upper hemicontinuity. Refinements like Reny’s “better-reply security” provide alternatives in discontinuous games.
• When single-valued, use Brouwer: If best responses are unique and continuous, Brouwer’s theorem suffices; invoking Kakutani is unnecessary overhead.

Related Concepts

Brouwer fixed-point theorem; Nash equilibrium existence; Debreu–Fan–Glicksberg fixed-point theorem; Tarski fixed-point theorem; Upper hemicontinuity and closed graph; Reny’s better-reply security.