Lawvere's fixed point theorem

Various diagonal arguments, such as those found in the proofs of the halting theorem, Cantor's theorem, and Gödel’s incompleteness theorem, are all instances of the *Lawvere fixed point theorem* (Lawvere 69), which says that for any cartesian closed category, if there is a suitable notion of epimorphism from some object $A$ to the exponential object/internal hom from $A$ into some other object $B$

$A \longrightarrow B^A$

then every endomorphism $f \colon B \to B$ of $B$ has a fixed point.

Let us say that a map $\phi: X \to Y$ is *point-surjective* if for every point $q: 1 \to Y$ there exists a point $p: 1 \to X$ that lifts $q$, i.e., $\phi p = q$.

**(Lawvere’s fixed-point theorem)** In a cartesian closed category, if there is a point-surjective map $\phi: A \to B^A$, then every morphism $f: B \to B$ has a fixed point $s: 1 \to B$ (so that $f s = s$).

Given $f: B \to B$, let $q: 1 \to B^A$ name the composite map

$A \stackrel{\delta}{\to} A \times A \stackrel{\phi \times 1_A}{\to} B^A \times A \stackrel{eval}{\to} B \stackrel{f}{\to} B$

from $A$ to $B$. In lambda calculus notation, $q = \lambda a: A. f \phi(a)(a)$. Let $p: 1 \to A$ lift $q$. Then calculate

$\phi(p)(p) = q(p) = (\lambda a: A. f \phi(a)(a))(p) = f \phi(p)(p)$

where the last equation is a beta-reduction. Hence $s \coloneqq \phi(p)(p)$ is a fixed point of $f$.

As pointed out by Lawvere, a hypothesis even weaker than point-surjectivity will do. Namely, $g: X \to Y^A$ is called *weakly point-surjective* iff for every $f: A \to Y$ there is $x: 1 \to X$ such that for every $a: 1 \to A$, we have $g(x)(a) = f(a)$. (This is weaker because $g(x) = f$ cannot be inferred from $g(x)(a) = f(a)$ for all *global elements* $a$.)

The statement need not hold if “(weakly) point-surjective” is replaced by “epimorphism”. For example, in the cartesian closed category of compactly generated Hausdorff spaces and continuous maps, with $S = S^1$ the circle, the Polish space $S^\mathbb{N}$ is compactly generated under the product topology; this is the exponential where $\mathbb{N}$ is given the discrete topology. There is a countable dense subspace $i: \mathbb{N} \to S^\mathbb{N}$, but recall that for any full subcategory of the category of Hausdorff spaces, a map $f: X \to Y$ is an epimorphism iff it has a dense image. On the other hand, there are obvious rotations of $S$ that have no fixed points.

Thus epimorphisms need not be (weakly) point-surjective. Nor are point-surjective maps necessarily epimorphisms; for example, if $U \hookrightarrow V$ is a proper inclusion between proper subobjects of the terminal object $1$ (as may happen in a sheaf topos), then this is vacuously point-surjective but not an epimorphism.

Point-surjectivity may seem like an inadequate notion of “epimorphism”, but it suffices for many purposes. For example,

**(Cantor’s theorem in a topos)** For any object $X$, there is an epimorphism $f: X \to \Omega^X$ only if the topos is degenerate.

Suppose there existed such an epi. In a topos, a map $f: X \to Y$ is epi iff the direct image map $\exists_f: \Omega^X \to \Omega^Y$ retracts the inverse image map $\Omega^f: \Omega^Y \to \Omega^X$, i.e., $\exists_f \circ \Omega^f = 1_{\Omega^Y}$. Putting $Y = \Omega^X$, the supposition implies that $\exists_f: Y \to \Omega^Y$ is a retraction. But retractions are automatically point-surjective.

We then conclude from Lawvere’s fixed point theorem that every endomorphism on $\Omega$, in particular the negation $\neg: \Omega \to \Omega$, has a fixed point $p: 1 \to \Omega$. Then $0 = p \wedge \neg p = p \wedge p = p$, whence $\neg 0 = 0$, or “true = false”: the topos is degenerate.

Another version of Lawvere’s fixed-point theorem requires only finite products for its statement. Namely, in a category with finite products, suppose $\Phi: A \times A \to B$ is a morphism with the property that for each $g: A \to B$ there exists $a: 1 \to A$ such that $g \lambda = \Phi \circ (a \times 1_A)$, where $\lambda: 1 \times A \stackrel{\sim}{\to} A$ is the projection. Then every map $f: B \to B$ has a fixed point. This version of the theorem is emphasized by Yanofsky.

Many applications of Lawvere’s fixed point theorem are in the form of negated propositions, e.g., there is no epimorphism from a set to its power set, or Peano arithmetic cannot prove its own consistency. However, there are positive applications as well, e.g., it implies the existence of fixed-point combinators in untyped lambda calculus.

In an interview (Lawvere 07) not long after Gödel’s 100th birthday, William Lawvere answered the question

We have recently celebrated Kurt Gödel’s 100th birthday. What do you think about the extra-mathematical publicity around his incompleteness theorem?

by saying (reproduced in Lawvere 69 reprint, p. 2):

“In ‘Diagonal arguments and Cartesian closed categories’ (Lawvere 69) we demystified the incompleteness theorem of Gödel and the truth-definition theory of Tarski by showing that both are consequences of some very simple algebra in the Cartesian-closed setting. It was always hard for many to comprehend how Cantor’s mathematical theorem could be re-christened as a ”paradox“ by Russell and how Gödel’s theorem could be so often declared to be the most significant result of the 20th century. There was always the suspicion among scientists that such extra-mathematical publicity movements concealed an agenda for re-establishing belief as a substitute for science. Now, one hundred years after Gödel’s birth, the organized attempts to harness his great mathematical work to such an agenda have become explicit.

The original article is

- William Lawvere,
*Diagonal Arguments and Cartesian Closed Categories*, Lecture Notes in Mathematics, 92 (1969), 134-145 (TAC)

A review and further development is in

- Noson Yanofsky,
*A Universal Approach to Self-Referential Paradoxes, Incompleteness and Fixed Points*, (arXiv:math/0305282)

Expositions include

- Andrej Bauer,
*On a proof of Cantor’s theorem*(web)

Revised on July 8, 2017 09:57:56
by Urs Schreiber
(188.100.200.251)