6
$\begingroup$

In the 2021 paper Level Theory Part I: Axiomatizing the Bare Idea of a Cumulative Hierarchy of Sets by Tim Button, a first order theory of the cumulative hierarchy is explored. Initially no axioms regarding the height of this hierarchy are added, resulting in a relatively weak theory referred to as ${\bf LT}$ for level theory. He then considers three more putative axioms regarding the height of this hierarchy, referred to as ${\bf Endless,\ Infinity}$ and ${\bf Unbounded}$. Towards the end of his paper we have

Proposition 7.1.

  1. ${\bf LT}$ proves separation, union and foundation.
  2. ${\bf LT + Endless}$ proves pairing and powersets.
  3. ${\bf LT+Endless+Infinity}$ proves Zermelo's axiom of infinity.
  4. ${\bf LT+Endless+\neg Infinity}$ is equivalent to ${\bf ZF_{fin}}$.
  5. ${\bf LT+Infinity+Unbounded}$ proves ${\bf Endless}$.
  6. ${\bf LT+Infinity+Unbounded}$ is equivalent to $ZF$.

These results are fascinating, but make me wonder if there is some elegant axiomatization of $ZF$ where remove the axioms of separation, union, and foundation (and perhaps pairing/powersets/more) and add some 'cumulative hierarchy' axiom and get the other axioms for free.

Is anyone here aware of such an axiomatization, or able to cook one up on the spot?

Another question that comes to mind, but which is slightly more ambiguous:

What is the axiomatic strength of the cumulative hierarchy over the rest of the set theory axioms, minus the ones we use to construct it?

Any pointers are appreciated.

$\endgroup$
12
  • $\begingroup$ I'm not sure about the extent to which this answers your question, but a lot of the axioms of ZF (union, replacement, infinity, power set, and even pairing if you cheat a bit) follow from the reflection scheme if formalized correctly. The approach I'm envisioning relies on foundation though. $\endgroup$ Commented Jun 25 at 20:47
  • $\begingroup$ Moreover, if reflection is stated strongly enough, separation can be replaced with $\Delta_0$-separation. $\endgroup$ Commented Jun 25 at 20:49
  • 1
    $\begingroup$ @AlecRhea: Is it possible to provide us the definitions of $\mathbf Endless$, $\mathbf Infinity$, and $\mathbf Unbounded$? It would help me ( and others) see how these three axioms are related as to the height of this hierarchy. Is there in fact a theorem in the paper you quoted that so relates these three axioms? Thanks in advance for supplying the requested info. $\endgroup$ Commented Jun 26 at 0:19
  • $\begingroup$ I'm not sure immediately how to answer @AlecRhea's question, but here is some context. LT automatically gives Union. I stated the axioms Endless, Infinity and Unbounded in terms of levels. The definition of "level" is itself a bit tricky, but we can give a few equivalents. $\endgroup$
    – Tim Button
    Commented Jun 26 at 12:27
  • $\begingroup$ Endless: Every level has a successor. Modulo LT, here are three equivalent formulations you could use instead: (1) Every set has a singleton. (2) Unordered-Pairing. (3) Powersets. $\endgroup$
    – Tim Button
    Commented Jun 26 at 12:28

2 Answers 2

6
$\begingroup$

This is an edited and improved answer; see edit details at the end.

We can obtain the whole of ZF using a single, natural, scheme.

I will keep the definition of level from Button 2021, as cited above. (The definition needs a slight tweak, because Button 2021 assumes extensionality, but that assumption is easily removed.) Intuitively, our scheme is as follows, for any definable partial function $\tau$. For any $a$, there is a level $s \ni a$ such that: for any $b \in s$, exactly one set in $s$ comprises the $\tau$-images of $b$'s members.

Here is the scheme precisely, with $Lev$ as in Button 2021:

The Main Scheme. Where $\tau$ is any definable partial function, this is an axiom: $$\forall a (\exists s \ni a)\big(Lev(s) \wedge (\forall b \in s)(\exists ! c \in s)\forall z(z \in c \leftrightarrow (\exists x \in b)\tau(x) = z)\big)$$

To avoid mis-understandings, here is how we might re-state the Main Scheme, in terms of arbitrary formulas. If $\forall x \forall y \forall z\big((\phi(x,y) \wedge \phi(x,z)) \rightarrow y = z\big)$, then $\forall a (\exists s \ni a)\big(Lev(s) \wedge (\forall b \in s)(\exists ! c\in s)\forall z(z \in c \leftrightarrow (\exists x \in b)\phi(x,z))\big)$.


Theorem. The theory whose only axioms are all the instances of the Main Scheme is a re-axiomatization of ZF.

To prove the Theorem, I will prove a few lemmas.

Lemma 1. The Main Scheme entails Separation. Proof. Fix $a$ and any formula $\psi(x)$. Let $\tau(x)$ be $x$ when $x \in a \wedge \psi(x)$, and be undefined otherwise. By the Main Scheme, there is a level $s$ with $a \in s$ and some (unique) $c \in s$ such that $\forall x\big(x \in c \leftrightarrow (x \in a \wedge \psi(x))\big)$. So $c$ is the result of Separation on an arbitrary set $a$.

Lemma 2. The Main Scheme gives Extensionality. Proof. Reason exactly as before, but letting $\psi(x)$ be trivial. So there is a level $s$ with $a \in s$ and a unique $c \in s$ which is co-extensive with $a$, i.e. $\forall x(x \in c \leftrightarrow x \in a)$. Moreover, if $d$ is co-extensive with $a$, then also $d \in s$, since $s$ is a level and hence is closed under subsets of members (see Button 2021 section 3). So all sets co-extensive with $a$ are in $s$, and hence are identical.

Note: having obtained Extensionality, we can re-write the Main Scheme like this, where $\tau$ is any definable partial function, and where we write $\tau[b]$ for $\{\tau(x) : x \in b\}$: $$\forall a (\exists s \ni a)\big(Lev(s) \wedge (\forall b \in s)\tau[b] \in s\big)$$ It's worth noting how this resembles Replacement.

Lemma 3. The Main Scheme gives Stratification and Endless. Proof. Let $\tau(x)$ just be $x$. By the Main Scheme, for any $c$ there is a level $s$ with $c \in s$. Now $c \subseteq s$ (as levels are transitive; see Button 2021 section 3) establishing Stratification. And also every level $c$ has a successor, establishing Endless.

Lemma 4. The Main Scheme gives Infinity. Proof. Let $\tau(x)$ be $\{x\}$. Using everything we have established so far, $\text{rank}(\tau(x)) = \text{rank}(x) + 1$ for every $x$. By Endless and the Main Scheme, there is some level $s$ with $\{\emptyset\} \in s$ such that: for any $b \in s$, also $\tau[b] \in s$. Since $\text{rank}(\tau[b]) = \text{lsub}_{x \in b}\text{rank}(x)$, it follows that $s$ is a limit level.

Lemma 5. The Main Scheme gives Unbounded. Proof. Let $\tau$ be any total definable function, i.e. $\tau(x)$ exists for any $x$. Fix $a$; by the Main Scheme, there is a level $s$ with $\tau[a] \in s$; now just note that levels are transitive, so $\tau(x) \in s$ whenever $x \in a$.

Combining Lemmas 1-5: the Main Scheme yields LT + Infinity + Unbounded, which is equivalent to ZF (as in Button 2021; the key to that equivalence is to prove, in ZF, that the $V_\alpha$'s are the levels). This final result completes the proof of our Theorem:

Lemma 6. ZF proves the Main Scheme. Proof. Fix $\tau$. Replacement tells us that $\tau(x)$ exists for any $x$. Using recursion, we can show that, for any $a$, there is a level $s$ with $a \in s$ and closed under $\tau[\cdot]$, i.e. if $b \in s$ then $\tau[b] \in s$.


Explanation of edits. In my original answer, I offered a simpler scheme: $$\forall b \exists c \phi(b, c) \rightarrow \forall a (\exists s \ni a)(Lev(s) \wedge (\forall b \in s)(\exists c \in s)\phi(b, c))$$ This a slight variation of a scheme suggested by Dana Scott (1960, "The notion of rank in set-theory"). The variation is suitable, given my definition of level.

Using arguments like those above (but simpler!), I showed that Extensionality + Separation + the Scott-like Scheme is just ZF under a different axiomatization.

Then @ZuhairAl-Johar suggested the improvement (in discussion below), tweaking the above to what I've now called the Main Scheme.

$\endgroup$
17
  • 1
    $\begingroup$ I think we can get matters to one axiom scheme on top of Extensionality. Given your definition of $Lev$ we can add that for any partial function symbol $f$, we have: $$ \forall a \exists s \ni a: Lev(s) \land \forall x \in s \exists y \in s: y=\{f(z)\mid z \in x\}$$. There is no need for Separation here. $\endgroup$ Commented Jul 4 at 10:11
  • 1
    $\begingroup$ Actually we can get matters down to ONE axiom scheme for all of $\sf ZF$. That is: if $f$ is a partial function symbol, then: $$\forall a \exists s \ni a: Lev(s) \land \forall x \in s \exists! y \in s: y=\{f(z)\mid z \in x\}$$. $\endgroup$ Commented Jul 4 at 16:38
  • 1
    $\begingroup$ @ZuhairAl-Johar: I think that's a very neat way to combine all of: Extensionality, Separation, Unbounded, Endless and Stratification. But I don't think it gives you Infinity; indeed, I think your scheme holds in the hereditarily finite sets. $\endgroup$
    – Tim Button
    Commented Jul 7 at 12:05
  • 1
    $\begingroup$ @TimButton, the singleton function will give you infinity. $\endgroup$ Commented Jul 7 at 12:07
  • 1
    $\begingroup$ I'll just quickly explain (for anyone else reading this) why @ZuhairAl-Johar's scheme gives Extensionality. Fix $c$, and let $f$ be this partial function: $f(x) = x$ if $x \in c$, but $f(x)$ is undefined otherwise. By the scheme: there is a level, $s$, with $c \in s$, and nothing else in $s$ is coextensive with $c$. But we must also check that there is no $d \notin s$ such that $d$ is coextensive with $c$. This holds because $s$, being a level, is closed under subsets, so that if $d$ is co-extensive with $c \in s$ then $d \in s$. $\endgroup$
    – Tim Button
    Commented Jul 7 at 12:13
2
$\begingroup$

Now, I think the bare idea of a Hierarchy is this:

if $F$ is a total unary function, then we can build a function $V$, such that:

$V^F_\emptyset = \emptyset \\ V^F_{\alpha+1} = F(V^F_\alpha) \\ V^F_\lambda = \bigcup V^F_{\alpha < \lambda}$

In the usual setting we take $F$ to be the powerset operator $\mathcal P$, but this is not the bare glimpse of a hierarchy. It is an instance of it.

However, since we want to capture $\sf ZFC$, then we can modify this to:

$V^F_\emptyset = \emptyset \\ V^F_{\alpha+1} = \mathcal P (F(V^F_\alpha)) \\ V^F_\lambda = \bigcup V^F_{\alpha < \lambda}; \text { for limit }\lambda$

When $F$ is the identity function, we'll simply omit $F$ from the notation.

Now, I think $\sf ZFC$ can be captured by defining all of the above and also defining the cardinality function, all in first order logic with equality and membership, then have axioms of

Extensionality: $x \subset y \land y \subset x \to x=y$

Levels: $\forall x \exists \alpha: |V^F_\alpha|=\alpha \land x \in V^F_\alpha$

Level-Separation: $\forall \alpha \forall \vec{k} \in V_\alpha \exists x: x=\{y \in V_\alpha \mid \phi^{V_\alpha}\}$

Where $\phi^X$ means all quantifiers in $\phi$ are bounded by $X$.

Lots of technical definitional details I didn't give here but they can be easily done. The point is that one can easily cook some hierarchy motivated principles from the outset, and get the resulting theory equivalent to $\sf ZFC$, but they would appear complex and in some sense contrived.

$\endgroup$
4
  • $\begingroup$ One could do things this way. But it makes a couple of assumptions I would prefer to avoid. (1) We have a recursion theorem which ensures the success of these definitions. (2) We explicitly want the stages of the hierarchy to be well ordered. I was—following in the footsteps of Dana Scott, and some others—considering how we might avoid building these two assumptions in at the outset. $\endgroup$
    – Tim Button
    Commented Jul 2 at 8:23
  • $\begingroup$ @TimButton, it depends on the definitions. If you define ordinals to be transitive sets of transitive sets not having more than one element with no successor, and so we'll not assume well ordering at the outset. As for the first point, I think it is circumvented by defining $x= V^F_\alpha$ in a piecemeal manner by the formula asserting the existence of a function $g$ from $\alpha+1$ with $g(\emptyset)=\emptyset$ and $g(m+1)=\mathcal P(F(m))$ and $g(\lambda)=\bigcup_{\beta <\lambda} g(\beta)$ for limit $\lambda$, and $x= g(\alpha)$. So, the symbol $V$ is eliminable totally. $\endgroup$ Commented Jul 2 at 10:42
  • $\begingroup$ I think you're assuming this all takes place within ZF(C). Again, there's no problem with that! But I wanted to work with much weaker assumptions; assumptions that can be justified by this story: Sets are formed at stages. At each stage, you find (all and only) every "possible" set whose members are among those things found earlier. Nothing seems to ensure the stages are well ordered, or to guarantee one can provide definitions by transfinite recursion. $\endgroup$
    – Tim Button
    Commented Jul 3 at 11:27
  • $\begingroup$ @TimButton, No, I'm not making this assumption either. I'm defining everything in the base language, then adding the axioms, then proving that ZFC would follow. I'm making no assumptions whatsoever other than those of first order logic with equality. I'm exactly starting from the base as you are doing. But, the formulation is more complex of course. $\endgroup$ Commented Jul 3 at 13:09

Not the answer you're looking for? Browse other questions tagged or ask your own question.