4
$\begingroup$

Fix a positive integer n. Assume $Lan=\{R_0,R_1,...,R_n\}$ be a language of first order logic, where every $R_i$ is a 2-ary relation symbol.

Assume $M$ is an Lan-model, where the underlying set is $ω_1$, and $\alpha R_0 \beta$ iff $\alpha<\beta$, and for $i>0$, $\alpha R_i \beta$ iff $L_\alpha$ is a $\Sigma_i$ elementary submodel of $L_\beta$.

Define $S\subset \omega_1$ be those ordinals $\alpha$ such that there is a finite Lan-model $A$ totally ordered by $R_0$, and $\alpha$ is the least ordinal such that there exists an Lan-monomorphism $A\rightarrow M$ such that the $R_0$-largest element of $A$ is mapped to $\alpha$.

Question: Calculate the order type of $S$.

$\endgroup$
8
  • 3
    $\begingroup$ This seems to be very much related to the calculation of the order type of Carlson's patterns of resemblance. And the latter question is open since the patterns have been introduced around 20 years. $\endgroup$ Commented Oct 25, 2022 at 10:12
  • 1
    $\begingroup$ @FedorPakhomov The story behind this question is interesting: in 2014 there was a contest to name large numbers where an anonymous contestant submitted a program (patched in 2018) claimed to take longer to terminate than Buchholz's hydra game, but its termination was only conjectural. Over the past couple years various amateur enthusiasts (including OP) have tried proving its termination and determining the size of its output, earlier this year OP made the connection that just by coincidence its algorithm looks similar to a variant of the patterns of resemblance comparison algorithm. $\endgroup$
    – C7X
    Commented Nov 3, 2022 at 2:19
  • $\begingroup$ @C7X Is there some link to this program? It would be curious to take a look. $\endgroup$ Commented Nov 4, 2022 at 14:51
  • 2
    $\begingroup$ @FedorPakhomov Here's a link to the 2018 patch of the program (the one most interested in this series of programs agree to be the canonical one), written in a dialect of BASIC. A word of warning: almost everything from the original entry to the 2018 version is conjectural, around 2016 it was found that the original 2014 program doesn't terminate on some inputs: it actually computed a non-well-founded notation whose well-founded part is at most $\Gamma_{\omega+1}$. No evidence of nontermination is yet shown for the 2018 version. $\endgroup$
    – C7X
    Commented Nov 4, 2022 at 23:25
  • 1
    $\begingroup$ Additionally, soon after a PDF (not peer-reviewed) was written up of a proof that the 2018 patch of the program terminates on inputs [[a,b],[c,d],[e,f],...] of two rows, using an order-preserving injection from this program's "hydra game" to Buchholz's 1984 ordinal notation for $\Pi_1^1-CA_0+BI$. The full program starts with input [[0,0,...,0,0],[1,1,...,1,1]] of length 10, so this is some progress. This PDF uses the notation (a,b)(c,d)... for [[a,b],[c,d],...], and it's in Japanese, I'm not sure if there's a translation. $\endgroup$
    – C7X
    Commented Nov 4, 2022 at 23:49

1 Answer 1

1
$\begingroup$

Edit 2024: I now think that this answer is wrong, specifically that the result in the last paragraph does not answer the original question. I am reluctant to delete this answer as it would also delete a helpful comment of Blass. I may update this answer in the future, however the question looks much more difficult when reading it correctly.

Fortunately I am not completely empty-handed: on p. 20 of Carlson's "Elementary Patterns of Resemblance" (Annals of Pure and Applied Logic vol. 108, 2001) there is a remark that gives an upper bound to this question with $Lan$ restricted to $\{R_0,R_1\}$ only, namely the proof-theoretic ordinal of $\Pi^1_1\mathrm{-CA}_0$.


There is a $\lambda<\omega_1$ such that $L_\lambda\prec L_{\omega_1}$, this can be obtained by taking the least ordinal not in $\textrm{Hull}^{L_{\omega_1}}(\{\varnothing\})$. The hull is already transitive (according to Marek and Srebrny's "Gaps in the Constructible Universe", or with a proof by Andreas Blass in the comments of this answer), so no Mostowski collapse is needed to put it in the form $L_\lambda$. Such $\lambda$ are unbounded in $\omega_1$: for any given $\xi<\omega_1$, by taking $\textrm{Hull}^{L_{\omega_1}}(\xi\cup\{\xi\})$ instead, a $\lambda_\xi>\xi$ can be produced such that $L_{\lambda_\xi}\prec L_{\omega_1}$.

Then, the ordinals $\alpha<\omega_1$ such that there is a $\beta<\alpha$ with $L_\beta\prec L_\alpha$ are unbounded in $\omega_1$. In particular, if $\beta,\alpha<\omega_1$ are such that $L_\beta\prec L_{\omega_1}$, $L_\alpha\prec L_{\omega_1}$, and $\beta<\alpha$, then $L_\beta\prec L_\alpha$. (This may be shown routinely by passing $\phi(\vec a)$ between $L_\beta$, $L_{\omega_1}$, and $L_\alpha$, with $\vec a$ a sequence of parameters from $L_\beta$.) This shows unboundedness since if some $\lambda_\xi$ is chosen for $\beta$, $\lambda_{\lambda_\xi}$ may be chosen for $\alpha$.

Since $\omega_1$ has cofinality $\omega_1$, this unbounded subset of $\omega_1$ must have order type $\omega_1$. Consider the finite $Lan$-model $\{\beta,\alpha\}$, where $\beta,\alpha$ are as in the last paragraph. $\beta R_n\alpha$ holds for all $n\in\mathbb N$, so to get a monomorphism to $M$ there would need to be a two-element subset $\{\gamma,\delta\}$ of $\omega_1$ such that $\gamma R_n\delta$ holds for all $n\in\mathbb N$. Taking $\gamma=\lambda_0$ and $\delta=\lambda_{\lambda_0}$, we get that any $\{\alpha,\beta\}$ is such a finite $Lan$-model. As such $\beta$ are unbounded in $\omega_1$, the order type of $S$ is $\omega_1$.

$\endgroup$
1
  • 1
    $\begingroup$ The Wikipedia statement has acquired a source, but it's easier to write a proof than to look up the source. Let $H$ be the hull in question, so $H\prec L_{\omega_1}$. Suppose $x\in y\in H$; we want to prove $x\in H$. Since $L_{\omega_1}$ satisfies "all sets are countable", so does $H$. So $H$ contains a function $f$ from $\omega$ onto $y$. Now $x=f(n)$ for some natural number $n$. Since $H$ contains both $f$ and $n$, it contains $f(n)$, as desired. $\endgroup$ Commented May 23 at 15:31

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .