📈 economics

There Ain't No Such Thing as a Free Equilibrium

The paper argues that the existence of universal equilibrium in games is simultaneously compatible and incompatible with the principle of avoiding strictly dominated strategies, depending on the specific sense in which these concepts are defined.

Original authors: Mark Whitmeyer

Published 2026-08-07

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Mark Whitmeyer

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to predict how people will behave in a game, from a simple game of Rock-Paper-Scissors to a complex, endless strategy war. This is the world of game theory, a branch of mathematics that studies how rational players make decisions when their outcomes depend on each other. The golden rule of this world is finding an "equilibrium"—a stable state where no one has a reason to change their strategy because they are already doing the best they can given what everyone else is doing. In small, finite games, we know these equilibria always exist. But when games get infinite, with endless choices, things get messy. To fix this, mathematicians sometimes use a special kind of math called "finitely additive probabilities." Think of this as a way of weighing options where you can assign a total weight of 100% to a whole group of choices, even if you assign 0% weight to every single individual choice inside that group. It's like saying a crowd is 100% full, even if every single person in it is invisible. The big question researchers have been asking is: Can we use this fancy math to find a perfect equilibrium in every game while also making sure we completely ignore "bad" moves? In game theory, a "strictly dominated" move is one that is always worse than another option, no matter what the opponent does. Common sense says a smart player should never play a dominated move, so a good solution should pretend those moves don't exist at all.

This paper, titled "There Ain't No Such Thing as a Free Equilibrium" (or TANSTAAFE), dives into a tricky conflict between two goals: finding an equilibrium in every possible game and ensuring that the solution completely ignores all the bad, dominated moves. The author, Mark Whitmeyer, proves that you can't have it both ways. If you want a solution that is guaranteed to exist for every game, you cannot demand that the solution completely ignores the entire set of bad moves at once.

Here is the twist: The paper shows that while you can make sure the solution ignores every single bad move individually, you cannot make it ignore the whole pile of bad moves together. Imagine a giant bag of rotten apples. You can easily point to one rotten apple and say, "I will not eat this one." You can do this for every single rotten apple in the bag. But the paper proves that in some infinite games, the "solution" might end up saying, "I will not eat any specific rotten apple," while simultaneously deciding to eat the entire bag of rotten apples. It's a paradox where the whole is treated as real, even though every part is treated as non-existent.

The author constructs a specific, tricky game to prove this point. In this game, there is an infinite list of moves that are all strictly worse than others. The paper demonstrates that any attempt to create a "perfect" solution that exists for all games and ignores the whole list of bad moves at once will fail; it leads to a mathematical contradiction. However, there is a silver lining. The paper shows that if you relax the rule just a tiny bit, you can still find a solution. Instead of demanding the solution ignores the whole bag of bad apples, you only demand it ignores each apple one by one. This weaker version works perfectly. The paper proves that there is a way to find an equilibrium in every bounded game where every single dominated move gets a "zero" rating, but the collection of all those moves might still get a "one" rating.

So, the main takeaway is a sharp boundary in the math of games. You can have a solution that exists everywhere and ignores every bad move individually, but you cannot have a solution that exists everywhere and ignores the entire group of bad moves as a single unit. The paper doesn't just suggest this; it provides a rigorous mathematical proof that such a "perfect" solution is impossible. It's a reminder that in the infinite world of game theory, sometimes you have to choose between having a solution at all and having a solution that behaves exactly the way our intuition says it should. The "free equilibrium" the title jokes about doesn't exist because you always have to pay a price: either you lose the guarantee of a solution, or you lose the guarantee that the solution completely dismisses the entire set of bad strategies.

Technical Summary: TANSTAAFE (There Ain't No Such Thing as a Free Equilibrium)

Problem Statement
The paper addresses a fundamental tension in game theory regarding the existence of equilibria in infinite games (games with infinite action sets). While finite games always possess Nash equilibria in mixed strategies, infinite games do not necessarily do so. A standard resolution involves expanding the strategy space from countably additive probabilities to finitely additive probabilities. While this restores equilibrium existence, it introduces a pathological distinction between "point-nullity" and "set-nullity" regarding strictly dominated actions.

Specifically, a finitely additive probability measure can assign probability zero to every individual action in an infinite set of strictly dominated actions (point-nullity) while simultaneously assigning probability one to the set of those actions as a whole (failure of set-nullity). The central question is whether a solution concept can simultaneously satisfy three desirable properties:

Universal Nonemptiness: Existence of a solution for every bounded game.
The Equilibrium Property: Adherence to coherence requirements for zero-sum games (specifically, the condition that the sum of evaluations of unilateral deviations does not exceed zero).
Set-Nullity: The requirement that the entire set of strictly dominated actions be ignored (assigned probability zero) by the solution.

Methodology
The author employs a functional analytic approach, utilizing the space of bounded real-valued functions $\ell^\infty(X)$ and evaluations (linear functionals) $\Lambda: \ell^\infty(X) \to \mathbb{R}$ .

Definitions: The paper defines a set $B$ as $\Lambda$ -null if altering payoffs on $B$ does not change the evaluation $\Lambda(B)$ . It distinguishes between point-nullity (every singleton dominated action is null) and set-nullity (the entire set of dominated actions is null).
Counterexample Construction: To test the compatibility of the three properties, the author constructs a specific bounded two-player zero-sum game, $\Gamma^*$ . In this game, the action sets are $C \cup \mathbb{N}$ (where $C=\{H, T\}$ and $\mathbb{N}$ is the set of natural numbers). The payoff structure is designed such that for every player, the natural numbers form a strictly increasing chain of dominance ( $n+1$ strictly dominates $n$ ), making the entire set $\mathbb{N}$ strictly dominated.
Finite Approximation: To establish positive results, the paper utilizes the concept of finitely approximable equilibria. This involves taking limits (in the weak-* topology) of mixed Nash equilibria from finite restrictions of the game, where the restrictions form an exhaustive net (eventually containing any finite subset of actions).

Key Results

Impossibility of Set-Nullity (Theorem 3.3):
The paper proves that no solution correspondence can simultaneously be universally nonempty, satisfy the equilibrium property for zero-sum games, and satisfy set-nullity.
- Mechanism: In the constructed game $\Gamma^*$ , if a solution satisfies set-nullity, the set of dominated actions $\mathbb{N}$ must be assigned probability zero. However, the payoff structure forces the evaluation of specific dominated actions (like action 1) to be positive when compared to the "safe" actions in $C$ . This leads to a contradiction with the zero-sum equilibrium condition (Condition Z), which requires $\Lambda_2(u_1) + \Lambda_1(u_2) \leq 0$ . Specifically, if $\mathbb{N}$ is null, the evaluations behave as if the players are restricted to $C$ , but the dominance structure implies that playing in $\mathbb{N}$ yields higher payoffs against certain opponent moves, creating an inconsistency that forces the solution set to be empty.
Feasibility of Point-Nullity (Theorem 4.7):
The paper demonstrates that a weaker requirement, point-nullity, is compatible with universal existence and the equilibrium property.
- Mechanism: By defining a solution correspondence $S_{pt}$ based on the marginals of finitely approximable equilibria (limits of equilibria of finite restrictions), the author shows that every strictly dominated single action is assigned probability zero. This is because, in any sufficiently large finite restriction, a strictly dominated action is never a best response and thus receives zero probability in the finite Nash equilibrium. The weak-* limit preserves this property for singletons.
Sharpness of the Distinction (Corollary 4.8):
The paper constructs a specific finitely approximable equilibrium for $\Gamma^*$ that satisfies point-nullity but fails set-nullity. In this equilibrium, every individual dominated action $n \in \mathbb{N}$ has probability zero, yet the set $\mathbb{N}$ as a whole has probability one. This confirms that the impossibility result for set-nullity is not an artifact of a specific construction but a fundamental limitation of the framework.

Significance and Claims
The paper claims a "sharp delineation" rather than a new equilibrium construction. Its primary contribution is establishing the boundaries of what is possible in infinite games:

Negative Result: It is impossible to demand that a universally existing equilibrium concept ignore the entire set of strictly dominated actions (set-nullity) while maintaining coherence in zero-sum games.
Positive Result: It is possible to demand that a universally existing equilibrium concept ignore each strictly dominated action individually (point-nullity).

The author argues that the pathology of finitely additive probabilities—where a set can have measure one while all its elements have measure zero—is unavoidable when dealing with infinite sets of dominated actions. The paper concludes that while finite approximation can eliminate fixed dominated actions, no universally existing solution can generally be required to ignore the entire, possibly infinite, set of dominated actions. This resolves the tension by showing that "free" equilibrium existence (without pathological side effects) comes at the cost of accepting that the set of dominated actions may not be collectively null, even if every individual dominated action is.

More like this