Category Theory
Zulip Server
Archive

You're reading the public-facing archive of the Category Theory Zulip server.
To join the server you need an invite. Anybody can get an invite by contacting Matteo Capucci at name dot surname at gmail dot com.
For all things related to this archive refer to the same person.

Stream: learning: questions

Topic: objectification tower termination (Lawvere fixed-point)

Giannis Karabesinis (Jul 27 2026 at 08:11):

I'm a non-mathematician working on a philosophy paper, and I've isolated a small purely mathematical question from it. I'd be grateful to know (a) whether the following is correct, and (b) whether it is already known / standard (I suspect it's essentially Lawvere + Tarski, but I want to be sure I'm not missing a subtlety or a reference).

Setup. A level is a triple $L=(A,Y,t)$ with $A$ an object of states, $Y$ an object of description-values, and $t:Y\to Y$ a distinguished endomorphism. The self-description space of $L$ is $Y^A$. Call $L$ self-complete if there is a weakly point-surjective $f:A\times A\to Y$ (every $g:A\to Y$ equals $f(a,-)$ for some point $a$). An objectification step sends $L$ to $L'$ whose state object contains a subobject $D$ with $d:D\times A\to Y$ representing every $g:A\to Y$, and which preserves $(Y,t)$.

Claim (characterization). Working in a category with finite products:

(1) Tower is open. If each level's value-object carries a fixed-point-free endomorphism preserved by objectification, then no level is self-complete, and there is no structure-preserving iso $L^*\cong \mathrm{objectify}(L^*)$.

Proof sketch. Contrapositive of Lawvere's fixed-point theorem. For the iso step: the iso fixes $(Y^*,t^*)$ and induces $A^*\cong A'$; since $d$ already represents maps out of $A^*$, relocating $D\subseteq A'$ along the iso yields $\tilde D\subseteq A^*$ with a representing $\tilde d$. I then assume $\tilde D$ is a retract of the state object ($e:\tilde D\to A^*$, $r:A^*\to \tilde D$, $r\circ e=\mathrm{id}$) and set $f(x,-)=\tilde d(r(x),-)$, which is weakly point-surjective; Lawvere then forces a fixed point of $t^*$, contradiction.

(2) Tower can close. The hypothesis is essential: in Scott domains ($D_\infty\cong[D_\infty\to D_\infty]$) and in Kripke's least-fixed-point theory of truth, self-completeness holds relative to the admissible class $\Delta$ (continuous / monotone maps), because the fixed-point-free operation (two-valued negation) is not admissible. Note the relativized reading of Lawvere here: the diagonal $q(x)=t(f(x,x))$ must itself lie in $\Delta$ to be represented, so the obstruction bites only when the fixed-point-free operation is itself in $\Delta$ — which is exactly why $D_\infty$ is consistent (two-valued negation isn't continuous, hence outside $\Delta$). Relative to $\Delta$: the tower closes $\iff \Delta$ contains no fixed-point-free endomorphism of $Y$.

My three specific uncertainties (where I'd stall):

1. Is the retract assumption in (1) the right/minimal hypothesis? My understanding is that retract is a clean sufficient condition giving both "points of $\tilde D$ lift to points of $A^*$" and "$f$ is total on $A^*\times A^*$" without a case-split; the minimal core seems to be just those two conditions (points lift + $f$ total), with the total-extension otherwise needing a complement (case-split, works in $\mathbf{Set}$/extensive categories). Is there a cleaner standard formulation?

2. The "$\Leftarrow$" direction of the characterization is solid. The "$\Rightarrow$" I can only state as "closure is then consistent and achievable" (via inverse-limit / final-coalgebra constructions), not that every $\Delta$ lacking the operation forces closure. Is a sharp "$\Rightarrow$" known, or is this the best one can do in general?

3. Is the whole thing just Lawvere + Tarski's undefinability + the Kripke revenge phenomenon repackaged? In particular, is the explicit phrasing "tower terminates $\iff$ no fixed-point-free endomorphism in $\Delta$" a recognizable result with a citable source, or does it not have a standard name?

Any pointers to existing literature would be very welcome. I know Lawvere 1969, Yanofsky 2003, Tarski, Kripke 1975, Scott 1972; happy to give the fully spelled-out proof (or a Lean version) if useful.

Oisín (Jul 27 2026 at 09:23):

Quickly scanning, I see some immediate errors. "Working in a category with finite products:" You need a Cartesian closed category to apply LFPT - this is already much stronger than finite products. You need this earlier too when you define the self-description space. I also don't understand why one would care about "weakly point-surjective" in any practical sense other than generality - point-surjectivity is more interesting.

Nathanael Arkor (Jul 27 2026 at 09:26):

Cartesian closure is not necessary for the fixed point theorem, e.g. see the introduction of @David Michael Roberts's Substructural fixed-point theorems and the diagonal argument: theme and variations.

Oisín (Jul 27 2026 at 09:27):

Nathanael Arkor said:

Cartesian closure is not necessary for the fixed point theorem, e.g. see the introduction of David Michael Roberts's Substructural fixed-point theorems and the diagonal argument: theme and variations.

Oh, thanks, I didn't know that! Sorry, I take it back.

Morgan Rogers (he/him) (Jul 27 2026 at 09:33):

I agree with you @Oisín that the definition of self-description space ostensibly needs cartesian closedness, though.

Giannis Karabesinis (Jul 27 2026 at 09:44):

Thank you both, this is very helpful.

@Oisín you're right writing "the self-description space is $Y^A$" presupposes an exponential, hence cartesian closure; that was a conflation on my part. What I actually intend is the finite-products formulation à la Yanofsky, working with a weakly point-surjective $A \times A \to Y$ and never forming $Y^A$ as an object. So the fix is to drop the $Y^A$ phrasing rather than to assume CCC.

That also answers your second question: weak point-surjectivity isn't there for its own sake , it's exactly what lets the diagonal run without exponentials, and for my application the non-cartesian-closed generality is the substantive point, not a technicality. (Point-surjectivity onto $Y^A$ would require $Y^A$ to exist, i.e. the very structure I'm trying not to assume.)

@Nathanael Arkor thank you for the Roberts reference — I didn't know it and it looks exactly on point. I'll study it before saying more; it seems directly relevant to whether the retract hypothesis is the minimal one, and to whether the "terminates $\iff$ no fixed-point-free endomorphism" phrasing is already known.

Nathanael Arkor (Jul 27 2026 at 09:50):

Perhaps also worth mentioning is another generalisation of Lawvere's fixed point theorem recently obtained by @Martín Hötzel Escardó.

David Michael Roberts (Jul 27 2026 at 09:50):

I will point out that Lawvere himself knew that cartesian closedness was not necessary and his two books formulate it or at least discuss it in the generality of only finite products. The papers by Yanofsky and myself come after these.

Matteo Capucci (he/him) (Jul 27 2026 at 09:55):

Giannis Karabesinis said:

An objectification step sends $L$ to $L'$ whose state object contains a subobject $D$ with $d:D\times A\to Y$ representing every $g:A\to Y$, and which preserves $(Y,t)$.

How does this step happen?

Notification Bot (Jul 27 2026 at 09:55):

This topic was moved here from #community: general > objectification tower termination (Lawvere fixed-point) by Matteo Capucci (he/him).

Giannis Karabesinis (Jul 27 2026 at 13:05):

Thanks — that's the right question, and my phrasing was misleading. I don't construct the next level: objectification is a relation, not an operation.

Definition. Say $L' = (A', Y', t')$ objectifies $L = (A, Y, t)$ if:

1. $(Y', t') = (Y, t)$ (the value object and its endomorphism are preserved);
2. there is a subobject $e : D \rightarrowtail A'$ and a map $d : D \times A \to Y$ such that every $g : A \to Y$ is $d(x, -)$ for some point $x$ of $D$;
3. $D$ is a retract of $A'$, i.e. there is $r : A' \to D$ with $r \circ e = \mathrm{id}_D$.

A tower is any chain $L_0, L_1, L_2, \dots$ with each $L_{n+1}$ objectifying $L_n$. I make no claim that such an $L'$ always exists, nor that it is unique — I only characterise what counts as such a step.

Claim. If $t$ is fixed-point-free, then no $L$ objectifies itself, and no level of a tower is self-complete. (Given (2) and (3), $f(x,-) := d(r(x),-)$ is weakly point-surjective $A' \times A \to Y$; Lawvere then forces a fixed point of $t$.)

So the "does the tower terminate" question becomes: can any $L$ objectify itself? Under a fixed-point-free $t$, no.

Kevin Carlson (Jul 27 2026 at 18:07):

Giannis, are you generating your responses via LLM? The first sentence of your last message strongly suggests that to my eye. Please don't do this.

Giannis Karabesinis (Jul 27 2026 at 18:25):

Yes I was using LLM so I can articulate my answers. I wasn't hiding it.

The ideas, the construction and also the mistakes are mine. I was using it to help me formulate my ideas as a  non-mathematician who wants to communicate clearly.

I am sorry that I did not disclose it from the beginning, it was the wrong move for this forum. From now on I will write everything by myself.

Kevin Carlson (Jul 27 2026 at 18:32):

Yes, it's perfectly understandable, it just makes it hard to be sure what you're really understanding, versus some mysterious superposition of you and the LLM. I think most of us here would agree that we'd rather see an unclear expression of your real state of understanding than the cleaner-looking LLM output.

Giannis Karabesinis (Jul 27 2026 at 18:36):

I understand thank you, i appreciate it, i will speak by myself from now on.

John Baez (Jul 27 2026 at 21:23):

Real people never say "Thanks — that's the right question, and my phrasing was misleading." :laughing:

David Michael Roberts (Jul 27 2026 at 21:29):

Whose phrasing was misleading, in this case? The LLM or the person trying to learn? If the latter, I have expensive experience teaching people 1-on-1 and trying to unpack their misconceptions and responding to their thinking process, so I know how to proceed. If the former, there is as far as I'm aware to theory of mind of LLMs, and it's a translation layer between me and the interlocutor, which makes things very very difficult

Giannis Karabesinis (Jul 28 2026 at 09:51):

Indeed there was an in-between LLM which helped me to articulate in mathematical language. The idea is purely mine. I was not trying to mislead anyone. Also the confusion was mine, the LLM just rendered my confused phrasing into formal language. I was trying to describe when a level stands in relation to the previous one but mistakenly I suggested a construction that produces the next level.As a non mathematician, I work on a philosophical problem and I would really appreciate your guidance.

Morgan Rogers (he/him) (Jul 28 2026 at 12:44):

@Giannis Karabesinis this is a strange definition, for the following reasons:

1. the endomorphism $t$ of $Y$ doesn't interact with the remaining structure. Perhaps you want to talk about $Y$ admitting a fixed-point-free endomorphism rather than including it in the structure?
2. The subobject $D$ of $A'$ is redundant. It seems to me that $(A',Y',t')$ "objectifies" $(A,Y,t)$ if and only if (point 1 holds and) there is a map $A' \times A \to Y$ such that every $g: A \to Y$ is of the form $d(x,-)$.

Taking the above into account (so dropping the $t$), I think the claim "if $t$ is fixed-point-free, then no $L$ objectifies itself" should read "if $Y$ admits a fixed-point-free endomorphism, then $L = (A,Y)$ does not objectify itself", with the proof indeed following from Lawvere's fixed point theorem. On the other hand, if $Y$ is a terminal object then $(A,Y)$ objectifies itself for any $A$ admitting a global element $1 \to A$.

Morgan Rogers (he/him) (Jul 28 2026 at 12:51):

If you want a more interesting example of a "self-objectifying" pair, you need to dig up some categories where the endomorphisms of objects have interesting fixed points AND you need to be careful about what count as elements in expressions like $d(x,-)$.

For instance, in the category of pointed sets, the two-element set $2$ has exactly two endomorphisms, and the pair $(2,2)$ "objectifies itself" if we consider the elements of these sets in the usual sense. BUT, these are not global elements in the categorical sense: a global element of $X$ is a function $1 \to X$, where $1$ is the terminal object, but because we're in the category of pointed sets, the terminal object is the set with a unique element (which is also the distinguished element), and there is only distinguished-element-preserving function $1 \to 2$ (or indeed to any set), so no function $A \times A \to Y$ can be pointwise surjective in the sense you defined!

Giannis Karabesinis (Jul 28 2026 at 19:50):

That helps a lot, thank you. I can see why taking out t from the construct makes everything simpler. And since the map A'×A→Y already catches every description, D and the retract are unnecessary.
I meant global element (1→A) when I said point and your pointed-sets example makes it clear now why that has to be stated.

Giannis Karabesinis (Aug 04 2026 at 12:29):

Following up from the cleanup of before, with t moved into the hypothesis and the retract/D dropped, i have arrived at a place in which i am unsure. A level is a pair (A, Y). Say (A', Y) objectifies (A, Y) when Y is preserved and there's a map d: A'×A → Y such that every g: A → Y arises as d(x, −) for some global element x: 1 → A'.

By Lawvere (contrapositive), if Y admits a fixed point free endomorphism then no level objectifies itself. What I want is a chain (A₀, Y) → (A₁, Y) → … where each Aₙ₊₁ objectifies (Aₙ, Y), and I would like to know if it it can ever terminate in a self-objectifying level.
There are two things I have a problem with.
a. Since the obstruction only sees Y and not A, is it right that Y admitting a fixed-point-free endomorphism once is enough to guarantee no level in the whole chain objectifies itself? That is the chain never terminates, and I don't need to recheck the condition at each step?
b. Is there any non-trivial situation where going from Aₙ to Aₙ₊₁ forces Y to change? For example, Aₙ₊₁ needs a richer codomain to represent the descriptions living over Aₙ? And if that can happen, could the new Y' fail to admit a fixed-point-free endomorphism, so this is the end of the ride and the chain actually stops?