Cohen overcounted: an existence proof

Update: This proof is flawed. I did not understand Boolean valued models. The sentence "P ranges over ℵ₂^M × ℵ₀", along with all other Th(M+P) sentences, is covered by the Boolean valued model, prior to forcing, via deductive closure. I will leave this failed proof up as a record of my intellectual journey.

This document proves that Cohen's ¬CH consistency proof is unsound. It does not refute Cohen's mathematics. It proves, by existence proof, that his construction counts extensions it has no right to count — and that it has no mechanism to tell the difference.

This is the second in a series. The first document verified the Ω Conjecture for all syntactically reachable cases and located the semantic limit where Woodin's Ultimate L program operates. This document removes the obstacle that has blocked that program from being taken seriously as a decision procedure for CH: the standard assumption that Cohen already settled the question by proving independence.

Cohen did not settle it. Here is the proof.

The setup

Let M be a countable standard transitive model of ZFC. Let P be Cohen's poset — the set of finite partial functions from ℵ₂^M × ℵ₀ to {0,1}, ordered by reverse inclusion. Let Th(M, P) denote the complete theory of M extended by P — the deductive closure of M ∪ P, everything provable once P is added to M as additional axioms.

This is the standard setup for Cohen's forcing construction. Nothing here is new.

The key observation: P is already defined inside M

Cohen's poset P is defined as:

Finite partial functions from ℵ₂^M × ℵ₀ to {0,1}, ordered by reverse inclusion.

The superscript M is not incidental. ℵ₂^M is the ordinal that plays the role of ℵ₂ inside M — it is M's cardinal, not an abstract placeholder. Cohen writes it into the definition of P directly. P is not a standalone object that is then combined with M. It is defined inside M from the start, with M's cardinal structure already baked into its domain.

This is a subtle but consequential sloppiness in Cohen's presentation. The standard picture of forcing treats M and P as separate ingredients — a ground model and a poset — that are combined to produce an extension. But Cohen's P is not independent of M. Its domain ℵ₂^M × ℵ₀ is supplied by M, and without M, P has no specific domain at all. Stripped of M, the schema "finite partial functions from some domain to {0,1}" is nearly contentless — a countable binary tree with no intrinsic connection to any cardinal.

The separation between ground model and forcing conditions that Cohen's framework presupposes is therefore already blurred at the point of definition. P's content is an interaction with M, visible in how P is written down in the first place.

Interaction sentences

Call a sentence φ an interaction sentence of Th(M, P) if:

The existence of interaction sentences is not surprising. Joint theories routinely prove things neither component proves alone. What matters here is that interaction sentences are outside the protection of the forcing relation.

The forcing relation — and the Truth Lemma that grounds it — protects explicit forcing conditions. If p ∈ P forces φ, then every M-generic extension containing p satisfies φ. This protection is exactly what the Truth Lemma guarantees. But interaction sentences are not forcing conditions. No condition in P forces them. They are visible to Th(M, P) but invisible to the forcing relation's checking mechanism.

The existence proof

φ: "P ranges over ℵ₂^M × ℵ₀"

This sentence is an interaction sentence of Th(M, P):

φ is therefore in Th(M, P) but is not a forcing condition.

Now consider the Boolean-valued model M^𝔹, where 𝔹 = RO(P) is the regular open algebra of Cohen's poset. Every sentence receives a Boolean value ‖φ‖ ∈ 𝔹. Since φ is not forced by any explicit condition p ∈ P, no condition p satisfies p ≤ ‖φ‖ = 𝟙. Therefore ‖φ‖ is strictly less than 𝟙 in 𝔹. Since φ is entailed by M and P jointly, ‖φ‖ is not 𝟘 either. Therefore:

𝟘 < ‖φ‖ < 𝟙

An intermediate Boolean value means exactly this: some M-generic ultrafilters U will have ‖φ‖ ∈ U, giving extensions where φ holds; and some M-generic ultrafilters U will have ‖φ‖ ∉ U, giving extensions where φ fails.

The extensions where ‖φ‖ ∉ U are forcing extensions that contradict a sentence of Th(M, P). They exist. The Boolean algebra witnesses it directly. □

Translation to Cohen's construction

The existence proof above uses the Boolean-valued model M^{RO(P)}. A natural question is whether the proof is an artifact of that reformulation — something visible only through the Boolean-valued lens, with no correspondent in Cohen's original construction. It is not. The translation is exact.

The connection between Cohen's poset P and the Boolean-valued model M^{RO(P)} is given by the canonical embedding:

i: P → RO(P) defined by i(p) = interior(closure(↑p))

This embedding satisfies two properties that matter here:

The forcing relation on P corresponds to Boolean values in RO(P) via:

p ⊩P φ iff i(p) ≤ ‖φ‖{RO(P)}

Now apply this to φ. In the Boolean-valued model, ‖φ‖ is strictly between 𝟘 and 𝟙. This means both ‖φ‖ and its complement ‖φ‖* are nonzero. By density of i(P) in RO(P), there exist actual conditions p, q ∈ P such that:

Translating back via the forcing correspondence: no single condition in P forces φ, and no single condition in P forces ¬φ. Cohen's own poset contains conditions on both sides. The gap is not introduced by the Boolean-valued reformulation. It is present in Cohen's original construction, pulled back from RO(P) to P via the density of the embedding.

The Boolean-valued model makes the gap visible. The embedding confirms it was always there. □

Cohen's construction has no mechanism to detect this

Cohen's construction selects a generic filter G by requiring it to meet every dense set in P. This guarantees that all explicit forcing conditions are respected — every finite partial commitment is honored, every method every implementation must provide is provided. The construction is sound with respect to explicit forcing conditions.

But dense set meeting says nothing about interaction sentences. The checking mechanism Cohen provides — the forcing relation, genericity, dense sets — is precisely the mechanism that protects explicit conditions. Interaction sentences are outside its scope by construction. No generic filter selection procedure based on dense sets can distinguish extensions where ‖φ‖ ∈ U from extensions where ‖φ‖ ∉ U, because φ is not a dense condition.

Cohen's construction therefore includes, among its ℵ₂^M many generic extensions, an unknown quantity of extensions that contradict interaction sentences of Th(M, P) — extensions where the domain of P is misread, where M's instantiation of the schema is not respected. Cohen has no way to identify which extensions are legitimate and which are not, because his construction provides no mechanism to check.

This is the precise sense in which Cohen overcounted.

Independence is not proven

The independence of CH rests on two pillars:

Gödel's pillar stands. V = L is a legitimate model, constructed by a canonical process that does not overcount.

Cohen's pillar is undermined. The existence proof above shows that Cohen's construction counts extensions that contradict interaction sentences of Th(M, P). Among the ℵ₂^M many reals Cohen adds, some come from extensions where ‖φ‖ ∉ U — extensions that misread the domain of P itself. Cohen cannot separate the legitimate extensions from the illegitimate ones. The model where CH fails may not be a valid model. The second pillar is unsound.

Therefore the independence of CH has not been proved. This is not a claim that CH is true or false. It is a precise logical observation: the standard argument for independence is incomplete.

What this means for Woodin

This result does not prove Woodin's Ultimate L program. It clears the path for it.

The standard objection to treating CH as an open question — rather than a settled independence result — is that Cohen already decided it. That objection is removed. CH is not independent. It is open.

Woodin's program constructs p_max, the maximal compatible forcing conditions, and asks what the genuine degrees of freedom are after all interaction sentences have been respected. If the genuine degrees of freedom collapse to ℵ₁, CH is true. The existence proof above shows that Cohen's count was not of genuine degrees of freedom — it was of all extensions, legitimate and illegitimate alike.

Together, the two documents in this series establish:

Neither document proves Woodin. Together they show he is looking in the right direction — which is the best that syntax can do.

The rest requires admitting the right extension. That is Woodin's work.