Tarski forbade the fixed point — here it is recovered, twice
A consistent language cannot contain a total, two-valued predicate that disquotes every one of its sentences: the Liar makes the fixed point Tr(⌜φ⌝) ↔ φ impossible to attain in full. But impossible in full is not impossible. Relax the demand that every sentence be true or false, and a language can hold its own truth predicate — as the least fixed point of a monotone operator, with the paradoxical sentences resting quietly in a gap. Relax instead the demand for a fixed point at all, and truth becomes a process of revision whose settled verdicts are the grounded sentences and whose eternal oscillation is the Liar. This study builds both, runs both, and reads them as the corpus's fixed-point construction recovered from the far side of its own impossibility.
after Tarski · the monotone operator · the least fixed point · the lattice · revenge · revision · the ground recovered
The Endorsement ended at a wall. A theory that would define its own truth runs into the Liar, and Tarski's theorem turns the collision into a proof: no consistent language has a predicate satisfying every instance of Tr(⌜φ⌝) ↔ φ. The classical response was to give up self-application — to stratify truth into a tower of languages, each speaking only of the one below. This study is about the other responses, the ones that keep a single self-applicable truth predicate and pay for it somewhere subtler than the hierarchy.
There are two, and each is a way of loosening what the fixed point demands. Kripke's, in 1975, loosens bivalence: it admits sentences that are neither true nor false, and shows that the truth predicate can then be built as an honest fixed point — the least fixed point of a monotone operator on partial interpretations, whose existence is the Knaster–Tarski theorem the corpus meets again and again. The paradoxical sentences do not break it; they fall into the gap it leaves, ungrounded. The revision theory of Gupta and Belnap loosens instead the demand for a fixed point: it keeps classical two-valued logic and the full biconditional, but reads the biconditional as a rule for revising a hypothesis about truth, iterated into the transfinite, and takes truth to be the pattern of what the revision stabilizes. We build the monotone operator and iterate it to its least fixed point; we survey the lattice of fixed points above it; we meet the revenge the gap invites; we run the revision sequence and watch the Liar flip forever while the grounded sentences settle. Two engines run live throughout, and on the sentences we examine both come to rest at the same verdict — the grounded truths settle, the Liar is left without one. The fixed point Tarski forbade is here — not whole, and not for free, but really, and in two independent hands.
Tarski's theorem is a prohibition with a precise shape: no consistent language holds a total, two-valued predicate disquoting all its sentences. The prohibition has exactly two joints, and loosening either — dropping totality-with-bivalence, or dropping the demand for a fixed point — lets the predicate back in, self-application and all.
Recall the wall. The diagonal lemma builds a sentence L with L ↔ ¬Tr(⌜L⌝); the disquotational schema Tr(⌜φ⌝) ↔ φ at φ = L gives Tr(⌜L⌝) ↔ L; and the two together force L ↔ ¬L. The proof uses three things at once — the schema holds of every sentence, every sentence is true or false, and the result must be consistent. Tarski's hierarchy keeps bivalence and consistency by surrendering the first: truth becomes level-indexed, and no predicate speaks of its own level. The theories of this study keep the predicate self-applicable and surrender one of the other two — silently, and only for the paradoxical sentences.
Admit sentences that are neither true nor false. The truth predicate becomes the least fixed point of a monotone operator; the Liar falls into the gap it leaves, neither in the extension nor the anti-extension.
Keep two-valued logic and the full biconditional, but read it as a rule for revising a hypothesis about truth. Iterate; truth is what the revision stabilizes. The Liar never stabilizes.
Both keep the one thing Tarski's tower could not: a single predicate that applies to the sentences of its own language, itself included. What each gives up is confined to where the paradox actually lives. We build the first now.
Kripke's move is to interpret the truth predicate partially — a set of sentences declared true, a set declared false, and the rest left open — and to close that interpretation under the obvious rule: a sentence is in the extension of truth exactly when it is true. That rule is a monotone operator, and monotone operators have least fixed points.
A partial interpretation of the truth predicate is a pair (E⁺, E⁻) of disjoint sets of sentences — the extension, declared true, and the anti-extension, declared false — with everything outside both left undefined. Order these by information: (E⁺,E⁻) ⊑ (F⁺,F⁻) when E⁺ ⊆ F⁺ and E⁻ ⊆ F⁻ — the second interpretation decides everything the first did, and perhaps more. Under this ordering the partial interpretations form a complete lattice.
Given a partial interpretation, evaluate every sentence by the Strong Kleene scheme — a conjunction is true if both conjuncts are true, false if either is false, undefined otherwise; a negation swaps true and false and fixes the undefined; and Tr(⌜φ⌝) takes the value the current interpretation gives φ. Then form the new interpretation
The decisive property is that adding information never retracts it. If (E⁺,E⁻) ⊑ (F⁺,F⁻), then anything the smaller interpretation evaluated true or false, the larger one evaluates the same way — the Strong Kleene scheme is monotone, undefined values only ever resolving, never flipping. So Φ is monotone: (E⁺,E⁻) ⊑ (F⁺,F⁻) implies Φ(E⁺,E⁻) ⊑ Φ(F⁺,F⁻). And that is all it takes.
The least fixed point is reached by iteration: start with the empty interpretation, apply Φ, and repeat. Sentences that bottom out in non-semantic fact acquire values and keep them — these are the grounded sentences. The Liar and its kin never acquire a value at all. Watch the construction run.
Begin with total ignorance — (∅, ∅), every sentence undefined — and apply Φ. Non-semantic facts resolve at once; a sentence asserting the truth of something already settled resolves at the next stage; and so the definedness spreads outward from the ground of fact. By monotonicity each stage extends the last, and the increasing sequence closes, at some ordinal, on the least fixed point. A sentence is grounded if it has a value there, ungrounded if it remains in the gap.
Grounding is well-foundedness of semantic dependence. A grounded sentence's truth traces back, in finitely or transfinitely many steps, to facts that do not mention truth at all — Tr(⌜0=0⌝) rests on 0=0, and Tr(⌜Tr(⌜0=0⌝)⌝) rests on that. The Liar rests on nothing but itself: to evaluate ¬Tr(⌜L⌝) one must already have evaluated Tr(⌜L⌝), which is L again. The iteration never gives it a first value to build on, so it stays undefined at every stage and in the fixed point. The paradox is not resolved and not suppressed; it is quarantined in the gap, where it can do no harm — and, as the next sections show, can still be pointed at from outside.
The least fixed point is one of many. The fixed points of a monotone operator themselves form a complete lattice, and the least one — grounded truth — is distinguished by deciding nothing that fact does not force. The structure separates two kinds of ungrounded sentence the single word “gap” had run together.
Knaster and Tarski proved more than the existence of a least fixed point: the fixed points of a monotone operator on a complete lattice themselves form a complete lattice, with a least and a greatest member. So Φ has a whole ordered family of fixed points — consistent interpretations at which truth and value coincide — and the least, the grounded one, sits at the bottom.
The lattice draws a line the least fixed point alone could not. The truth-teller τ, saying only Tr(⌜τ⌝) — this sentence is true — is undefined in the least fixed point, exactly as the Liar is. But τ is not paradoxical: there is a fixed point in which it is true and another in which it is false, each perfectly consistent. It is merely unfounded — nothing settles it, so grounded truth declines to, but either choice would cohere. The Liar is different in kind: it is in the gap of every fixed point, for no consistent interpretation can place it in the extension or the anti-extension. Groundlessness and paradox are not the same defect — the truth-teller floats free of fact; the Liar is at war with it — and only the lattice of fixed points, not the least point by itself, tells them apart.
Kripke's language keeps consistency by admitting gaps — and pays in expressive power. Its negation reaches only as far as falsity; the notion it most needs, not true in the sense that includes the gap, is precisely the notion it cannot express of itself. Say that notion, and the Liar returns.
In the Strong Kleene scheme, ¬Tr(⌜φ⌝) is true when φ is false and undefined when φ is undefined. So the object language's “not true” means “false,” not “untrue,” and there is no predicate in it for the third status — no way to say, inside the language, that a sentence is gapped, or equivalently that it is not true in the wider sense of “false-or-gapped.” That predicate belongs to the classical metalanguage that surveys the fixed point from outside.
Now form the sentence that uses exactly the missing notion: this sentence is not true, with “not true” read as “false or gapped.” If it is true, it is not true; if it is false or gapped, then it is not true, so it is true. The contradiction is back — written in the one word the repair had to step outside the language to obtain. Kripke's construction is genuinely consistent, but its consistency is bought by an expressive gap, and Tarski's hierarchy reappears at the level of that gap: to talk about which sentences are gapped is already to have climbed to a metalanguage the object language cannot contain. The deficit of The Endorsement is conserved; the gap is only its new address.
The other route refuses the gap. It keeps classical two-valued logic and the whole biconditional, and pays the price on the opposite side of the ledger — by giving up the fixed point. The biconditional becomes a rule for revising a guess about truth, and truth is taken to be whatever the revision settles down to. The grounded sentences settle; the Liar oscillates forever.
Start from a hypothesis h — a complete, two-valued guess at the extension of the truth predicate, right or wrong. Use it to evaluate every sentence classically, and then revise the guess by the disquotational rule: in the next hypothesis, Tr(⌜φ⌝) holds exactly when φ held under the old one. The revised hypothesis δ(h) is a better guess — it has corrected the truth predicate to agree with the biconditional one step further out. Iterate δ, into the transfinite, and watch what stabilizes.
The verdict of revision is read off the pattern. A sentence is categorically true if, from every starting hypothesis, the revision stabilizes it at true after some stage — and these categorically stable sentences are what the theory calls true. Every grounded sentence of Kripke's construction is categorically stable in revision, and on the sentences in play here — the settled facts, the truths raised upon them, the Liar, the truth-teller — the two theories, built from opposite materials, a partial fixed point against a classical process with no fixed point, return the same verdict. Where they diverge is only on the ungrounded: Kripke gives them a single verdict, the gap, while revision distinguishes the merely unstable from the periodic, and the transfinite limit rules of Herzberger and Belnap fix what happens at limit stages, where a sentence stable below the limit keeps its value and an unstable one is reset. Truth, on this account, is not a state a language is in but a pattern a process falls into — and the Liar's pattern is to have none.
The corpus reads every limitative result as a diagonal forbidding a fixed point. Truth is the semantic case: Tarski's theorem forbids the fixed point of the truth operator in full. This study has watched that fixed point recovered — once by holding it in a weaker logic, once by denying it and keeping the pattern of its pursuit — and both recoveries turn on the same machinery the corpus finds everywhere.
Set the two constructions against the impossibility they answer. Tarski: no total, bivalent Tr with Tr(⌜φ⌝) ↔ φ everywhere. The two escapes each keep the biconditional and self-application, and each relaxes one further thing.
Drop bivalence. Tr is the least fixed point of a monotone operator — the Knaster–Tarski fixed point the corpus meets in every guise. The paradox rests in the gap; grounded truth is what fact forces.
Drop the fixed point. Tr is revised, not solved; classical logic is kept throughout, and truth is the categorically stable pattern of the revision. The Liar's pattern is to oscillate.
Tarski's theorem is usually read as a closed door: a language cannot hold its own truth. It is better read as a door that opens only sideways. Insist that every sentence be true or false and the door stays shut; loosen that, and truth walks back in as the least fixed point of a monotone operator, the paradoxes resting in the gap it leaves, grounded truth resolving outward from the ground of fact. Insist instead on a fixed point and the door stays shut; loosen that, keep classical logic entire, and truth walks back in as the settled pattern of an endless revision, the grounded sentences coming to rest and the Liar flipping without end. The two constructions share their materials with the rest of the corpus — a monotone operator and its least fixed point, a diagonal sentence that no assignment can satisfy — and they share their verdict on the sentence that matters: the truths that the ground of fact reaches are held, and the one sentence that reaches only for itself is not. What the language cannot hold, on either account, is the Liar. That sentence is not destroyed and not admitted. It is left where the fixed point leaves it — in the gap, or in the oscillation — the diagonal's refusal made survivable at last.