Determinacy (ZF + AD)

Origin. Mycielski and Steinhaus, "A mathematical axiom contradicting the axiom of choice" (1962), proposed AD and showed it refutes AC. Solovay, Martin, and Moschovakis developed the structure theory through the 1970s; Martin (1975) proved Borel determinacy in ZF, and (1970) Π¹₁-determinacy from a measurable. Martin and Steel (1985/1989) derived projective determinacy from infinitely many Woodin cardinals; Woodin proved the converse direction and the theorem that L(ℝ) ⊨ AD.

Models. ZF with choice replaced by a hypothesis about infinite games. Two players alternate playing naturals; the payoff is a set A ⊆ ℕ^ω; AD says every such game is determined — one player has a winning strategy. The axiom is false in ZFC, since a well-ordering of ℝ builds an undetermined set by transfinite recursion, and it is not a weakening of choice but a competitor: under AD every set of reals is Lebesgue measurable, has the Baire property, and has the perfect set property, so the pathologies of ZFC's real line all vanish at once. The price is exactly AC, and the consistency price is exactly infinitely many Woodin cardinals.

Formalism.

Language: ∈. The signature is ZF's; the axiom is new.

The game G_A: players I and II alternate, I playing a₀, II playing a₁, I playing a₂, …, each aᵢ ∈ ℕ. The play is x = ⟨a₀, a₁, a₂, …⟩ ∈ ℕ^ω. I wins iff x ∈ A.

Axiom of Determinacy (AD): for every A ⊆ ℕ^ω, G_A is determined — I has a winning strategy or II does. The theory is ZF + AD; DC (dependent choice) is usually added, and AD implies countable choice for sets of reals.

AD refutes AC: Given a well-ordering of the 2^ℵ₀ strategies, recurse: at stage α, defeat the α-th strategy for I by putting a play into A, and the α-th for II by keeping one out. The resulting A is undetermined. Two lines, and they use choice essentially.

What AD gives (Mycielski–Świerczkowski, Banach–Mazur, Davis, Morton–Solovay): Every set of reals is Lebesgue measurable — no Vitali set. Every set of reals has the Baire property — via the Banach–Mazur game. Every set of reals has the perfect set property — so CH holds in the form "every set of reals is countable or has size 2^ℵ₀", with no intermediate cardinality possible. ω₁ is measurable. ω₁ and ω₂ are measurable in ZF + AD. There is no uncountable well-ordered sequence of distinct reals; ℝ is not well-orderable.

Consistency strength (Woodin): Con(ZF + AD) ⟺ Con(ZFC + there exist infinitely many Woodin cardinals). This is the exact calibration, and it is the reason AD is taken seriously rather than dismissed as a curiosity: the two hypotheses arose from unrelated motives and land on the same point.

L(ℝ) ⊨ AD (Woodin): from ω Woodin cardinals with a measurable above, the inner model L(ℝ) satisfies AD. So one need not adopt AD as true of V; it is true somewhere, in a canonical inner model containing every real, and the structure theory applies there. This is how AD is actually used — as a theory of L(ℝ) under large cardinals, not as an alternative foundation.

The determinacy ladder, by definability class: Open/closed (Gale–Stewart 1953): ZF. Borel (Martin 1975): ZF; and Friedman showed it needs ℵ₁-many iterations of the power set — provable in ZFC, not in ZF−P. Analytic Π¹₁ (Martin 1970): a measurable cardinal. Π¹ₙ, projective (Martin–Steel): n Woodin cardinals with a measurable above; projective determinacy from infinitely many. AD^{L(ℝ)} (Woodin): ω Woodins with a measurable above. Full AD in V: inconsistent with AC, hence never adopted in V by anyone using choice.

Stronger variants: AD_ℝ (games on reals rather than naturals) — strictly stronger, implies AD; AD⁺ (Woodin), the technical strengthening under which the structure theory is developed and which is not known to differ from AD.

Symbols.

Metatheory. The equiconsistency with infinitely many Woodins is what makes ZF + AD a theory rather than a provocation. AC and AD are both consistent relative to ZF and they contradict each other, so one might expect the pair to be symmetric in the way AC and ¬AC are — Gödel's L gives Con(ZF) → Con(ZFC), Cohen's forcing gives Con(ZF) → Con(ZF + ¬AC), and neither costs anything. AD costs infinitely many Woodin cardinals. That measures exactly how much stronger AD is than a mere denial of choice, and the measurement was possible only because the Woodin hierarchy had been developed for unrelated reasons. Placement here follows the criteria: the signature is ZF's, the axiom is not valid in the base logic, and dropping it changes the theorems. The division has Borel Determinacy cross-listed from Applications/Game as a theorem and Large Cardinals in Metatheory as the hierarchy; neither is the theory, and the theory is what descriptive set theory works in.

Applies to. Descriptive set theory — the entire regularity-property structure of the projective hierarchy is a consequence of determinacy at the corresponding level. Inner model theory, through L(ℝ) and the derived model theorem. The calibration of large cardinal strength: determinacy hypotheses are the standard yardstick, and the correspondence is exact rather than approximate. Infinite game theory, Banach–Mazur and related games. The philosophy of set theory, as the standard case of a hypothesis with strong extrinsic justification and no intrinsic one.

Limitations. Inconsistent with AC, so ZF + AD is not a foundation for ordinary mathematics: no basis for every vector space, no Hahn–Banach in general, no Tychonoff. This is why it is used as a theory of L(ℝ) under large cardinals rather than as an axiom of V, and stating it as an alternative to ZFC misdescribes the practice. Its consistency is not provable from ZFC, or from anything ZFC-users are entitled to — infinitely many Woodins is a substantial assumption. The structure theory is usually developed under AD⁺, whose relation to AD is not settled. The full axiom's status as true of any natural universe has no defender; what it has is a calibration.

© 2026 Lingenic LLC