Deep apparatus · III — the instances, worked

The Fates

This is where the claim is tested. The corpus says that Cantor and Russell, Gödel and Tarski, the halting problem, Curry's paradox, and the recursion theorem are not analogues but instances — the one construction of The Fulcrum, read at different four-tuples (A, B, φ, α). Here each is worked in full: the actual Gödel diagonal lemma with its substitution function, Turing's diagonal program built by the s-m-n theorem, Curry's explosion without a single negation, Kleene's recursion theorem proved by the same self-application. And running through all of them is a single finding — that the forced point s = α(s) can meet exactly four fates, and which one is decided by the status of the naming and the behaviour of the twist.

(A, B, φ, α)↦g↦s = α(s)↦fate

A unification earns its name only if the things unified are really the same thing. It is cheap to say that the diagonal arguments "resemble" one another; the corpus says more — that they are one argument, the construction g = α ∘ ev ∘ ⟨φ, 1⟩, differing only in the four-tuple at which it is read. That is a checkable assertion, and this document checks it, instance by instance, by exhibiting for each the category, the index object A, the value object B, the naming φ, and the twist α, and then watching the same g produce the same forced equation s = α(s).

What the survey reveals is not a uniform outcome but a structured one. The forced point is sometimes delivered — recursion, a program that knows its own code. Sometimes it is forbidden — the naming that would produce it is shown not to exist, and a non-existence theorem falls out of consistent mathematics. Sometimes it is explosive — the naming was granted by axioms, so the impossible point becomes a contradiction and the system collapses. And sometimes it is undecided — the naming is provability, weaker than truth, and the gap between them absorbs the twist into incompleteness rather than contradiction. Four fates; one construction; and a procedure that names the fate from the data. We state the procedure, then run it.

1 · the template

One construction, four destinies — the machine that reads any instance

Every instance is the same five lines at a different four-tuple. The outcome is governed by two questions only: whether the naming φ genuinely exists, and whether the twist α has a fixed point. Their answers partition every diagonal phenomenon into four.

Recall the construction from The Fulcrum. Given a category with finite products, a naming φ : A → Bᴬ, and a twist α : B → B, form g = α ∘ ev ∘ ⟨φ, 1⟩ : A → B — the map that feeds each index its own named value, twisted by α — and if φ reaches g on points, the value s = g(a) at the naming index satisfies s = α(s). Whether that equation is a gift or a catastrophe depends on the data, and the dependence is completely systematic.

the two governing questions

Q_α — does the twist have a fixed point? If yes, the construction lands in the power face: s = α(s) is satisfiable and is handed over. If no, it lands in the limit face: s = α(s) is impossible, and something must give.

Q_φ — what is the standing of the naming? In the power face this is moot. In the limit face it decides everything: a naming genuinely present cannot have a fixed-point-free twist (that is the power case); a naming merely hypothesized from outside is refuted (it cannot exist); a naming posited by axioms turns the impossibility into a theorem (the system explodes) — unless the posited predicate is only one-directional, provability rather than truth, in which case the contradiction is deferred into incompleteness.

Those answers carve the landscape into four fates. Each is a real region, populated by real theorems.

benignα has a fixed point ·
φ genuinely surjects

The point is delivered. The naming is a true point-surjection — a reflexive object, or the computable universe with its universal machine — so the power face applies and every twist has a fixed point. This is recursion: a transformation of programs that is met by a program transformed into itself.

Kleene's recursion theorem · the Y combinator · quines · Rice's theorem (as corollary)

absentα fixed-point-free ·
φ hypothesized externally

The naming is forbidden. A surjection, a decider, a definition is supposed for contradiction; the impossible s = α(s) shows it cannot exist. No axiom is harmed — a non-existence theorem simply falls out of consistent mathematics.

Cantor (no surjection onto ℘) · Turing (no halting decider) · Tarski (no truth predicate)

explosiveα fixed-point-free ·
φ posited by axioms (full biconditional)

The system collapses. The naming is granted — unrestricted comprehension, an unrestricted predicate — as a full biconditional. The impossible point is then a theorem, and the theory proves a contradiction. Inconsistency, not non-existence.

Russell's paradox · Curry's paradox (no negation required)

undecidedα = ¬ · φ = provability,
one-directional only

The contradiction defers into a gap. The naming is provability — a genuine predicate, but only Prov(⌜φ⌝) → φ, never the converse. The diagonal sentence is therefore independent, true but unprovable, rather than contradictory. The distance between ⊢ and ⊨ is exactly what it absorbs.

Gödel's first incompleteness · Gödel's second (via Löb)

the fate decision procedure — answer the questions; the fate of the forced point follows

The same construction every time; only the four-tuple changes, and these three questions read its fate from the data. The tree is not a heuristic — each branch is a theorem worked in the sections below. That a single decision procedure sorts Cantor from Gödel from the recursion theorem is the precise content of "one construction."

2 · the absent

Forbidden namings — Cantor, Turing, Tarski

Three of the most famous impossibility theorems are one event: a naming is hypothesized, the twist is fixed-point-free, and the forced equation refutes the naming. No axiom is touched; the impossibility is a theorem of consistent mathematics.

In each case below a rich naming is supposed — a surjection onto a power set, a halting decider, a truth definition — and the supposition is shown to collapse the moment the diagonal is run against a fixed-point-free twist. The casualty is always the naming itself.

Cantor — no set surjects onto its power set

Take A any set, B = {0,1}, α = ¬, and a hypothesized surjection φ : A → ℘(A) = Bᴬ. The construction's g(x) = ¬[x ∈ φ(x)] is the diagonal set D = {x : x ∉ φ(x)}.

(1)D = { x ∈ A : x ∉ φ(x) } ∈ ℘(A); by surjectivity D = φ(a) for some a.

∎a ∈ φ(a) = D ⟺ a ∉ φ(a). Contradiction. No surjection exists: |℘(A)| > |A|.α = ¬ is fixed-point-free, so D is named by no point — the prototype of every absent fate.

The next is Cantor transported into the category of computable functions — with one subtlety that turns out to be the difference between consistency and paradox.

Turing — the halting problem is undecidable

Work in the partial computable functions: A = ℕ (program codes), B = {↓, ↑} ≅ {halts, loops}, α = swap. The universal machine makes e ↦ φ_e a genuine — but partial — naming. Suppose the total halting predicate H(e, x) = [φ_e(x)↓] were computable; then the construction's g is a program D that inverts the diagonal: D(x) = loop if H(x,x) = 1, else halt. By the s-m-n theorem D is a program with a code d.

∎D(d) ↓ ⟺ H(d, d) = 0 ⟺ φ_d(d) = D(d) ↑. Contradiction. So H is not computable.The diagonal program loops exactly when it halts — Cantor's anti-diagonal, made to run.

why partiality is the difference between recursion and paradox

The computable universe is reflexive enough to be dangerous: the universal machine is a real point-surjection of codes onto partial functions, and that is precisely what powers the recursion theorem of §5. What it is not is a point-surjection onto total two-valued maps — for a total halting decider H would upgrade it to exactly that, and α = swap forbids it by Cantor's argument. The single fact that programs may diverge is what keeps the universe consistent while leaving it self-referential. Remove partiality and the recursion theorem and the halting paradox would demand the same impossible object. Self-reference survives only because it is allowed to loop.

Tarski — truth is not definable

In a theory T ⊇ Q, suppose a formula True(x) satisfied the T-schema — T ⊢ φ ↔ True(⌜φ⌝) for every sentence φ — making it a naming of A = sentences onto B = {⊤, ⊥}. Take α = ¬. The diagonal lemma (§4) supplies the construction's fixed point L — the Liar — with T ⊢ L ↔ ¬True(⌜L⌝).

(1)T ⊢ True(⌜L⌝) ↔ L (T-schema at φ = L).

(2)T ⊢ L ↔ ¬True(⌜L⌝) (diagonal lemma at α = ¬).

∎T ⊢ True(⌜L⌝) ↔ ¬True(⌜L⌝), so T ⊢ ⊥. As T is consistent, no such True(x) exists in its language.A truth predicate would be a point-surjection onto the two truth values; α = ¬ forbids it exactly as it forbids Cantor's surjection.

Three theorems, one shape. A naming is hypothesized from outside the system; the fixed-point-free twist makes the forced point impossible; the naming is refuted. Cantor loses a surjection, Turing a decider, Tarski a definition — and in every case the ambient mathematics stays consistent, poorer by exactly one object it could never have had. The next fate is what happens when the system was unwise enough to grant the naming as an axiom.

3 · the explosive

Granted namings — Russell, and Curry without negation

When the system's own axioms assert the rich naming, the impossible point is no longer hypothetical — it is a theorem, and the theory proves a contradiction. Curry's version sharpens the lesson: the explosion needs no negation, only the rule of contraction that is the diagonal.

The absent fate cost mathematics nothing — a naming was supposed and refused. The explosive fate is the same construction with the naming built in. Unrestricted comprehension grants, for every predicate, a set that collects exactly its instances; that is a posited point-surjection of the universe onto its own predicates, and the diagonal turns it into ruin.

Russell — naive comprehension is inconsistent

Naive comprehension posits, for every ψ, a set {x : ψ(x)} — making membership a granted naming onto B = {⊤, ⊥}. Take α = ¬; the construction's g is Russell's set R = {x : x ∉ x}.

(1)By comprehension, ∃R ∀x ( x ∈ R ↔ x ∉ x ). Instantiate x = R.

∎R ∈ R ↔ R ∉ R, so the theory proves ⊥. It is inconsistent — not merely missing an object, but collapsed.α = ¬ has no fixed point, yet comprehension forces one to exist. The impossible point is a theorem.

One might think negation is the culprit — that paradox needs the destructive twist. Curry's paradox removes it, and shows the explosion is structural, not logical.

Curry — explosion with no negation at all

Fix any sentence Y whatsoever. Comprehension at ψ(x) = (x ∈ x → Y) grants a set C with x ∈ C ↔ (x ∈ x → Y). The twist is α = (· → Y), and C is the construction's forced fixed point. Writing c for C ∈ C, comprehension gives c ↔ (c → Y).

(1)Assume c. From c ↔ (c→Y), the assumption gives c → Y; with c again, modus ponens yields Y. Discharge: c → Y.The hypothesis c is used twice — once to obtain c→Y, once as the antecedent. This is contraction.

(2)From c ↔ (c→Y) and (1): c.

∎From (1) and (2), modus ponens gives Y — for every Y. The theory proves everything. No negation appears anywhere.Taking Y = ⊥ recovers Russell; the general Y shows the catastrophe is triviality, driven by contraction alone.

the rule that giveth and taketh away — contraction, again

Curry's derivation used its hypothesis twice. That is the structural rule of contraction — and contraction, The Fulcrum showed, is exactly the diagonal Δ, the comonoid copying that makes self-reference expressible. So one rule stands at both ends of the phenomenon. Contraction is what lets Lawvere's g duplicate its input and exist at all; and contraction is what lets a posited self-referential naming detonate, by using the forced fixed point twice to prove anything. The substructural logics that drop it — linear, affine, the Grishin and Girard naive set theories — block Curry's paradox and can carry unrestricted comprehension without contradiction. They pay for it precisely where one would expect: with no diagonal, they also have no recursion theorem. The capacity to copy is indivisible; it grants the construction and the collapse in one stroke.

The boundary between absent and explosive is the standing of the naming, nothing else. Hypothesize a surjection and lose it (Cantor); assume an axiom that grants it and lose the system (Russell, Curry). Between the two sits the most delicate fate of all — where the naming is real, but too weak to either be refuted or to explode.

4 · the undecided

The gap that absorbs the diagonal — Gödel

Tarski's truth predicate exploded because truth is a full biconditional. Gödel's naming is provability — a real predicate, but only one-directional. That single weakness converts the diagonal from a contradiction into an independent sentence, and incompleteness is the result.

This fate is the subtlest because the naming is genuine. Provability is a definable arithmetic predicate; nothing is hypothesized and nothing is granted illegitimately. What makes the difference is that provability implies truth but truth does not imply provability — the predicate is one-directional — and so the diagonal sentence it produces is not forced into contradiction. It is merely forced out of reach. First the engine that has powered every instance, now proved.

the diagonal lemma — the syntactic form of weak point-surjectivity

For every formula θ(y) in the language of a theory T ⊇ Q, there is a sentence G with T ⊢ G ↔ θ(⌜G⌝). Every definable property of sentences has a sentence asserting that property holds of itself. This is the self-naming that Lawvere's theorem requires, realized in the category of formulas by the representability of substitution.

proof — the substitution function, made to name itself

The diagonalization function diag(⌜ψ(x)⌝) = ⌜ψ(⌜ψ(x)⌝)⌝ — substitute a formula's own code into its free variable — is computable, hence representable in T by a formula Diag(x, y) with T ⊢ Diag(⌜ψ⌝, y) ↔ y = ⌜ψ(⌜ψ⌝)⌝. Now set ψ(x) := ∃y\,(Diag(x, y) ∧ θ(y)) and define G := ψ(⌜ψ⌝). Then

∎G = ψ(⌜ψ⌝) ⟺ ∃y(Diag(⌜ψ⌝, y) ∧ θ(y)) ⟺ θ(⌜ψ(⌜ψ⌝)⌝) = θ(⌜G⌝).The only step is that Diag(⌜ψ⌝, y) pins y to ⌜ψ(⌜ψ⌝)⌝ = ⌜G⌝. Self-reference is substitution, made representable.

the provability predicate and its one weakness

Prov_T(y) := ∃p\,Prf_T(p, y), where Prf_T(p, y) represents "p codes a T-proof of the sentence y" — a Σ₁ predicate. It satisfies the derivability conditions: (D1) T ⊢ φ ⟹ T ⊢ Prov(⌜φ⌝); (D2) provability respects modus ponens; (D3) T ⊢ Prov(⌜φ⌝) → Prov(⌜Prov(⌜φ⌝)⌝). The decisive fact is what is absent: the converse of soundness, Prov(⌜φ⌝) ↔ φ, is not available — that would be a truth predicate, which §2 forbade. Provability runs one way only.

Gödel's first incompleteness theorem

Apply the diagonal lemma at θ(y) = ¬Prov_T(y): a sentence G with T ⊢ G ↔ ¬Prov_T(⌜G⌝) — "G says it is unprovable." This is Lawvere's construction in the Lindenbaum category, with naming φ = Prov and twist α = ¬.

(1)If T ⊢ G: then by (D1) T ⊢ Prov(⌜G⌝); but T ⊢ G → ¬Prov(⌜G⌝), so T ⊢ ⊥. By consistency, T ⊬ G.

(2)Since G is unprovable, ¬Prov(⌜G⌝) holds in the standard model — so G is true.

∎If T is ω-consistent then T ⊬ ¬G (Rosser's device weakens this to bare consistency). So G is independent — true, yet neither provable nor refutable.The forced point did not contradict; it stepped outside what T can decide.

why this is incompleteness and not paradox — the ⊢/⊨ gap, exactly

Set Gödel beside Tarski. Tarski's naming obeyed the full biconditional True(⌜φ⌝) ↔ φ, so the Liar L ↔ ¬True(⌜L⌝) became True(⌜L⌝) ↔ ¬True(⌜L⌝) — an outright contradiction, the explosive fate. Gödel's naming is Prov, and the biconditional Prov(⌜φ⌝) ↔ φ is unavailable: provability gives truth, but truth need not give provability. So G ↔ ¬Prov(⌜G⌝) never collapses to Prov(⌜G⌝) ↔ ¬Prov(⌜G⌝). Instead it describes a perfectly consistent situation — G true and unprovable — and the diagonal, finding no contradiction to make, makes an independent sentence instead. The entire distance between incompleteness and paradox is the distance between ⊢ and ⊨: a one-directional naming has slack, and the diagonal falls into the slack. Tarski has none, and the diagonal detonates.

Gödel's second incompleteness theorem, via Löb

If T is consistent, then T ⊬ Con(T). Löb's theorem — itself a diagonal result — states T ⊢ (Prov(⌜φ⌝) → φ) ⟹ T ⊢ φ. Take φ = ⊥: then Prov(⌜⊥⌝) → ⊥ is exactly ¬Prov(⌜⊥⌝) = Con(T), so T ⊢ Con(T) ⟹ T ⊢ ⊥. A consistent T cannot prove its own consistency. The fixed-point construction that gives G gives Löb's theorem too — and the modal logic GL in which Löb's principle is the defining axiom is the subject of the next deep-dive, the refinement.

The undecided fate is the construction at its most refined: a real naming, a real twist, and a forced point that neither arrives nor contradicts but hovers — independent. It exists only because provability is honest about being weaker than truth. Where the naming claims more than it can deliver (Tarski, Russell) the diagonal punishes the overreach; where it claims exactly what it can (Gödel) the diagonal merely marks the limit. One fate remains, and it is the only happy one.

5 · the benign

The point delivered — recursion, quines, and Rice

When the naming is a true point-surjection — the computable universe, with its universal machine — the power face applies and every twist has a fixed point. The construction stops forbidding and starts building: this is recursion, and its proof is Lawvere's g with the reflexive structure made real.

Every fate so far has been a "no" — no surjection, no decider, no truth, no consistency, no decision. The benign fate is the single "yes." Here the naming genuinely exists: the universal machine makes every partial computable function nameable by a code, an honest point-surjection. Lawvere's theorem in its forward direction then guarantees that any transformation of programs is met by a program it sends to itself.

Kleene's second recursion theorem

For every total computable f, there is an index e with φ_e = φ_{f(e)}. A program that behaves identically to its own transform under f — a fixed point of f up to extensional equality. The category is the partial computable functions, A = ℕ, the naming φ the universal machine, and α = f an arbitrary index transformation.

(1)By s-m-n, a total computable d with φ_{d(x)} ≃ φ_{φ_x(x)} for all x (run x on x to get an index, then run that).The self-application φ_x(x) is the diagonal — "x applied to x", the categorical ⟨id,1⟩.

(2)Given f total, f ∘ d is total computable; let v be an index for it: φ_v = f ∘ d. (v is the name ⌜g⌝.)

∎Set e := d(v). Then φ_e = φ_{d(v)} ≃ φ_{φ_v(v)} = φ_{(f∘d)(v)} = φ_{f(d(v))} = φ_{f(e)}.e = d(v) is g(a); the self-application v↦φ_v(v) is exactly the WW of the Y combinator. This proof is Lawvere's, in the computable category.

this proof is the Y combinator, and these are quines

The shape is unmistakable: d encodes self-application, v names the transform, and e = d(v) closes the loop — precisely g, ⌜g⌝, and s = g(a) of The Fulcrum, with φ_v(v) the WW that defines Y. The recursion theorem is Lawvere's theorem read forwards on the one genuinely reflexive object mathematics hands us for free. Its most concrete fruit: take f(e) to be an index for the program "print the string e." The fixed point e then satisfies φ_e = "print e" — a program that prints its own source. Every quine is a fixed point of the diagonal, delivered.

The benign fate even reaches back to manufacture absent ones. A delivered fixed point, dropped into a decision problem, forces a program into a class it was sworn to avoid — and the avoidance becomes impossible.

Rice's theorem — a benign fixed point yields an absent decider

Let S ⊆ ℕ be semantic (membership depends only on φ_e) and non-trivial (∅ ≠ S ≠ ℕ). Then S is undecidable.

(1)Suppose S decidable. Fix a ∈ S, b ∉ S. Define f(e) = b if e ∈ S, else a (computable, since S is decidable).

(2)By the recursion theorem, ∃e with φ_e = φ_{f(e)}.

∎If e ∈ S then φ_e = φ_b, so e ∉ S (semantic, b ∉ S) — and dually. Either way a contradiction; S is undecidable.The recursion theorem builds a program e that imitates b when classed with a, and a when classed with b: no decision can hold.

The benign fixed point is the construction's gift, and Rice shows it is not idle: recursion, the only happy fate, is strong enough to forbid every non-trivial decision about programs. The gift and the bar are, once more, the same hand. With all five fates in view, the survey can be laid out at once.

the instance inspector — one construction, seven readings; the four-tuple is the only thing that changes

Each instance is the same g = α ∘ ev ∘ ⟨φ, 1⟩ at a different (A, B, φ, α), producing the same s = α(s). The category, the naming, and the twist are the dials; the fate is the reading. That seven landmark theorems are rows of one table — not a family resemblance but a literal substitution of parameters — is the whole of what the corpus means by one construction.

Seven theorems, one construction, four fates. The diagonal set and the diagonal program, the Liar and Russell's set, the Gödel sentence, the Curry trap, the self-printing quine — each is g at a four-tuple, each forces s = α(s), and each meets its destiny by the standing of its naming and the temper of its twist. Delivered, forbidden, explosive, undecided: the four exhaust the ways a thing can be made to meet itself. The construction is one; what differs is only what we ask of it, and what it is allowed to answer.