The Mirror · movement IV of VI
The fates ask what the forced point becomes when nothing in the values will seat it. This movement asks the prior question they all presume: by what right does any fixed point exist at all? The answer is one resource worn three ways — and the deepest turn in the corpus, where the diagonal’s own arena is revealed to be a limit, and the two mechanisms collapse into one.
A fixed-point theorem is the claim that some completeness is strong enough to defeat some class of twist. Mathematics runs on exactly three — and the limits are what the strongest of them costs.
The fates asked what the forced point becomes when no value will hold it. The supply side asks the prior question they all presume: by what right does a fixed point exist at all? Mathematics answers in three idioms, and beneath the three lies a single resource. A fixed point is a place where a map stops moving; to be sure such a place exists, you need a space complete enough that the search for it cannot fall through the floor. Each genus is one species of that completeness.
Completeness of naming. The object holds a name for every one of its own maps into B. So total a mirror forces a fixed point of every twist at once — which is exactly why it alone can fail. Lawvere, Cantor, Gödel, Turing.
Completeness of order. Every set of approximations has a least upper bound. A monotone map then has a least fixed point — the limit of ⊥ ≤ f⊥ ≤ f²⊥ ≤ ⋯. The order does the work; self-reference is never asked. Denotational recursion; Kripke's truth.
Completeness of convergence. Every Cauchy sequence has a limit; every open cover, a finite subcover. A contraction or continuous self-map then has a fixed point — the limit of its own iterates. Nash equilibria; contractions; reflective oracles.
Each genus is a real theorem with a real proof, and setting the three out exactly shows what the word completeness does in each — and where the order theorem parts company with the diagonal one.
| theorem | completeness | map class | the fixed point |
|---|---|---|---|
| Knaster–Tarski | complete lattice | monotone | μf = ⋀{x : fx ≤ x}, νf = ⋁{x : x ≤ fx} |
| Kleene | pointed dcpo | continuous | μf = ⊔ₙ fⁿ(⊥) |
| Banach | complete metric | contraction q<1 | unique x* = lim fⁿ(x₀) |
| Brouwer | compact convex | continuous | some x* — no construction |
The order theorem is the one to prove in full, because the corpus leans on it twice. Let L be a complete lattice, f monotone, and P = {x : f(x) ≤ x} the pre-fixed-points. Put a = ⋀P. For every x ∈ P we have a ≤ x, so f(a) ≤ f(x) ≤ x; thus f(a) is a lower bound of P, giving f(a) ≤ a — that is, a ∈ P. Apply f once more: f(f(a)) ≤ f(a), so f(a) ∈ P and hence a ≤ f(a). The two inequalities give f(a) = a, and as the meet of P it is the least fixed point. ∎
Notice what the proof did not use: continuity, or any notion of iteration. Knaster–Tarski seats the point by a single meet, for any monotone map. Kleene’s ⊔ₙ fⁿ(⊥) is a sharper, narrower instrument — it demands the map be continuous, so that it commutes with the supremum, but then delivers the fixed point constructively, as the limit of a computable orbit. The gap between the two theorems is precisely the maps that are monotone but not continuous: there the ascending orbit ⊥ ≤ f⊥ ≤ f²⊥ ≤ ⋯ exhausts ω without closing, and one must climb transfinitely — fα(⊥) through the ordinals, taking suprema at limit stages — until a closure ordinal γ with fγ+1(⊥) = fγ(⊥) is reached. Knaster–Tarski guarantees the destination; the transfinite orbit is the journey to it. That is the order genus entire, and it is the resource the rest of this movement keeps spending.
Set the three beside one another and a pattern sharper than "completeness" appears: the genera divide by mechanism. Two of them find their fixed point by convergence — iterate the tame map and its orbit closes on a limit, ⊔ fⁿ(⊥) in a complete order, lim fⁿ(x) in a complete metric — so what must be complete is the orbit, and a fixed point is seated only for maps tame enough to converge. The third finds its fixed point by diagonal: no iteration, no limit, only the mirror reading its own name back through the twist. What must be complete there is the representation — the object names its own endomaps — and because that asks nothing whatever of the map, it hands a fixed point to every map at once. Both are completeness. They are completeness of different things — and that difference is exactly where limitation is born.
That asymmetry is the precise origin of limitation. A convergence theorem can never contradict itself: a map that will not converge is simply handed no fixed point, and nothing is claimed against it. The diagonal theorem can contradict itself, because it has already promised a fixed point to a map that demonstrably has none — negation on the two truth-values. So the classical limits are not some fourth, defective genus, and not a failure of completeness at all. They are the diagonal mechanism colliding with the one twist its own generosity cannot honor. The same overreach that makes the mirror powerful — a fixed point for everything — is what makes it impossible wherever a single twist refuses to rest.
Here is the corpus compressed to a sentence. A fixed-point theorem is a completeness theorem. Order-completeness defeats the monotone twists; metric and topological completeness defeat the continuous ones; representational completeness — the mirror — defeats all twists, and so is the only completeness that can turn on its owner. Recursion is what that representational completeness grants when the twist can be seated; limitation is the same completeness meeting a twist that cannot be — ¬ on the two truth-values, whose sole escape is that no mirror onto them exists. The genera are not three unrelated theorems. They are two mechanisms for one phenomenon — fixed by limit, or fixed by mirror — and limitation lives in the mirror, the only mechanism whose reach is wide enough to indict itself.
This sharpens into a clean criterion for which worlds can even be limited. The contrapositive bites on an answer-space B exactly when B carries one endomap with no fixed point. On the two truth-values that map is unique — it is negation — which is why crisp, two-valued logic is precisely the soil in which the diagonal forbids. At the far pole sits a reflexive object D ≅ Dᴰ: there every map has a fixed point, supplied uniformly by Y, so no twist can be fixed-point-free and the mirror is never contradicted. Power and limitation are the two ends of one spectrum, and the only question the world answers is whether its answers leave room for a map that refuses to rest.
Their effects differ — only the diagonal overreaches — but their resources are one theorem at two heights, and seeing this is the deepest turn the corpus takes. The diagonal's arena, a reflexive object D ≅ [D → D], is itself a fixed point — not of a map but of the functor X ↦ [X → X] — and fixed points of functors obey the law fixed points of maps do, one storey up. Lambek's lemma says an initial algebra is already such a fixed point, F(μF) ≅ μF; Adámek's theorem builds it as the colimit of the chain 0 → F0 → F²0 → ⋯. This is Knaster–Tarski with the word poset rewritten category: where the order theorem climbs ⊥ ≤ f⊥ ≤ f²⊥ ≤ ⋯ to the least fixed point of a monotone map, the categorical one climbs 0 → F0 → F²0 → ⋯ to the initial fixed point of a functor — and Scott's D∞ is exactly that climb for the function-space functor, its chain D₀ → [D₀→D₀] → ⋯ resolving, by the limit–colimit coincidence, into D∞ ≅ [D∞ → D∞]. So representational completeness is no rival to order-completeness; it is order-completeness, lifted from the lattice of points to the category of objects. There were never two resources — only one construction, the colimit of an orbit, run at the level of points to yield Kleene's value and at the level of objects to yield Scott's mirror, inside which the diagonal then turns.
The callout’s claim — that a reflexive object is a fixed point of a functor, obeying the laws fixed points of maps obey — is itself two theorems, and proving them is the precise sense in which the diagonal’s arena is built by the order mechanism.
The proof is pure initiality. (F A, Fα) is also an F-algebra, so there is a unique algebra map h : (A, α) → (F A, Fα); and α itself is an algebra map (F A, Fα) → (A, α). Their composite α ∘ h is an endomorphism of the initial algebra, so by initiality α ∘ h = idA. Then h ∘ α = F(α ∘ h) = F(id) = id. So h = α⁻¹, and α is iso. ∎ Dually, a final coalgebra A → F A is an isomorphism too — the greatest fixed point.
This is Kleene’s theorem with every symbol promoted one storey: the initial object 0 for the bottom ⊥, the functor F for the continuous map, the colimit for the supremum, Fⁿ0 for fⁿ(⊥). The algebra structure is exactly the witness that F preserved the colimit: F(colim Fⁿ0) ≅ colim Fⁿ⁺¹0 ≅ colim Fⁿ0. Run it on the function-space functor F(X) = [X → X] and you are building Scott’s D∞. The one technical obstruction — that [X→X] is contravariant in its first slot, so the chain is not a plain diagram of maps — dissolves in the category of CPOs and embedding–projection pairs, where Smyth and Plotkin’s limit–colimit coincidence makes the colimit of the embeddings equal to the limit of the projections: a single bilimit D∞ satisfying D∞ ≅ [D∞ → D∞]. So the reflexive object inside which the diagonal turns is, demonstrably, an order-theoretic limit. There were never two resources. The mirror is the colimit, seen from within.
The initial algebra has a twin, and the pair sharpens what completeness grants. Order-completeness offers not one fixed point but two: a least, built upward from ⊥ — the inductive μ, the realm of finite, well-founded, terminating things — and a greatest, built downward from ⊤, the coinductive ν, the realm of streams, processes, and computations that never end. Terminating recursion lives in the first; the non-settling ⊥ that the recursion fate hosts lives in the second. Scott's reflexive object is special precisely because there the two coincide — it is bifree, initial algebra and final coalgebra at once — which is why it can model the whole untyped calculus, the halting and the diverging together. And this coincidence is the exact knife-edge of the discipline wall. A fixed-point operator at every type, applied to the identity at the empty type, yields an inhabitant of ⊥ — a proof of falsehood. To admit Y everywhere is to be Turing-complete and inconsistent; to forbid it is to be a consistent logic that cannot name every program. Curry–Howard makes the trade literal: non-termination is inconsistency, and a total type theory purchases its soundness with precisely the self-reference it declines.
The bifree coincidence is not a curiosity; it is the exact location of the discipline wall, and the boundary can be drawn with one term. Suppose a fixed-point combinator YA : (A → A) → A at every type A. Take A = ⊥, the empty type — False under Curry–Howard. The identity id : ⊥ → ⊥ always exists, so Y⊥(id) : ⊥ is an inhabitant of falsehood, and by ex falso every type is then inhabited. General recursion at every type is logical inconsistency — not analogous to it, identical under propositions-as-types.
The converse is Tait’s theorem (1967): the simply-typed λ-calculus is strongly normalizing — every term reaches a normal form in finitely many steps — and strong normalization forces consistency, since ⊥ has no normal inhabitant and therefore none at all. The two facts close the knife-edge to a point. A calculus with Y at every type is Turing-complete and inconsistent; a strongly normalizing one is consistent and total but cannot express every computable function. The universal fixed-point operator is the line between them — the very line §III drew between recursion and incompleteness, redrawn in the calculus of proofs. To hold Y is to compute everything and prove nothing sound; to refuse it is to prove soundly and compute less. The discipline wall is that refusal, written as a typing rule.
Now every escape the corpus records reads as one move: when representational completeness meets a twist it cannot seat, re-host that twist where a weaker completeness can. Drop negation into the three-valued lattice {T, F, ⊥} and it turns monotone; order-completeness seats it; the Liar comes to rest at ⊥ — Kripke's truth. Drop the reflexive set Cantor forbids into Scott's domains and self-application turns continuous; the same order-completeness returns it as a least fixed point — Scott's D ≅ Dᴰ, the home of recursion. The point always comes back. But the bill is exact, and it is paid in the completeness surrendered: you leave the crisp, total, two-valued world the mirror ruled and accept truth-value gaps, partial maps, a continuity restriction — a change in what your objects are. You never defeat the diagonal. You emigrate to a country with looser laws.
And the looser country has a diagonal of its own. Every recovery enriches the language — to park the Liar at ⊥ you must be able to say "undefined"; to seat self-application you must be able to name the new objects — and a richer language is a richer mirror. The twist tamed at one level returns, sharpened, one level up. Walk the three escapes and watch the obstruction come back each time it is paid off.
So the two ledgers close into one. Read as demand, the forced point is absent, divergent, paradoxical, or undecided — four ways the values fail to hold it. Read as supply, it is held after all, by order- or by metric-completeness, at the price of leaving the world that forbade it. Every limit in the corpus is a fixed point some completeness refused to seat; every recovered power is the same fixed point seated by a completeness elsewhere, on terms. There is one forced point — and the whole corpus is the accounting of where it is allowed to land.
A movement of The Mirror, the keystone of Quodlibet. The theorem and its five-line proof are Lawvere’s (1969); the synthesis is the author’s.