This is a dual question to my previous question, here: Can two first-order theories agree on finite structures but disagree on infinite structures?. My current question is, does there exist a first-order signature $L$ and two first-order $L$-theories $T$ and $T'$ such that both $T$ and $T'$ have infinite models and also arbitrarily large finite models, and the infinite models of $T$ and $T'$ are the same, but the finite models of $T$ and $T'$ are not the same? Also, bonus question, if the answer is yes, is it possible for either or even both of $T$ and $T'$ to be finitely axiomatizable?
-
3MSE is not a "do my (home)work for me" site. Posts like this (e.g. "PSQs") are in poor taste, & often downvoted & closed; we expect users to put in & show your effort! If you want help, [edit] your post to add relevant context (more info here), e.g. relevant definitions, your work/attempts, the problem's source, where exactly you're stuck, what you do/don't understand & have learned recently, etc. Also: this is useful for formatting LaTeX, & Approach0 helps find past questions. – PrincessEev Feb 12 '25 at 18:02
-
4@PrincessEev I'm not sure your comment is useful for a user with 23k rep who has been using the site (and asking questions exactly like this one) for more than 11 years. I'm not defending the question or saying people shouldn't cast downvotes and close votes. But I'm sure user107952 is aware of site policies and has received comments like yours many times in the past. – Alex Kruckman Feb 12 '25 at 18:09
-
1There are easy examples, e.g. $(\exists x, \exists y, (x\neq y))\to \forall x, P(x)$ and $\forall x, P(x)$. These two sentences have the same infinite models, but they are not equivalent, since the first has a model of size $1$ in which the unique element fails to satisfy $P$. Can I suggest a more interesting variant? Are there theories $T$ and $T'$ which have the same infinite models, both have arbitrarily large finite models, and for all $n\in \omega$, $T$ and $T'$ disagree about models of size $\geq n$? – Alex Kruckman Feb 12 '25 at 18:18
-
@Alex: hmm. In your variant what happens if you take the ultraproduct of an increasing family of models that the theories disagree about? – Qiaochu Yuan Feb 12 '25 at 18:29
-
2@QiaochuYuan We could have a sequence $(M_i)_{i\in \omega}$ such that each $M_i$ is a model of $T$ but not a model of $T'$, and yet the ultraproduct $(\prod M_i) / \mathcal{U}$ is a model of both $T$ and $T'$. Your ultraproduct argument does rule out having finitely axiomatizable $T$ and $T'$ as in my comment (since for finitely axiomatizable $T$, the class of non-models of $T$ is closed under ultraproducts too). – Alex Kruckman Feb 12 '25 at 19:23
-
This question is perfectly clear and natural to ask – there are a number of important theorems about first-order logic concerning the cardinalities of models. I'm voting to reopen. – Joe Feb 14 '25 at 03:15
1 Answers
Converting my comments into an answer:
There are easy finitely axiomatizable examples, e.g. $T = \{(\exists x \exists y\, (x\neq y))\rightarrow P\}$ and $T' = \{P\}$. [Here $P$ is a proposition symbol, i.e., a $0$-ary relation symbol, which is set to either true or false in a structure. If you're not comfortable with proposition symbols, you can replace $P$ with $\forall x\, P(x)$ everywhere in this answer.]
These two sentences have the same infinite models, and have arbitrarily large finite models, but they are not equivalent, since the first has a model of size $1$ in which $P$ is false.
A more interesting variant is the following: Are there theories $T$ and $T'$ which have the same infinite models, both have arbitrarily large finite models, and for all $n\in \omega$, $T$ and $T'$ disagree about models of size $\geq n$?
The answer is no if we additionally require $T$ and $T'$ to be finitely axiomatizable. Suppose for contradiction we have finitely axiomatizable $T$ and $T'$ as above. For all $n\in \omega$, we can pick a finite structure $M_n$ with $|M_n|\geq n$ and either $M_n\models T$ and $M_n\not \models T'$, or $M_n\models T'$ and $M_n\not \models T$. By passing to a subsequence, without loss of generality we can assume that $M_n\models T$ and $M_n\not\models T'$ for all $n\in \omega$. Now let $\mathcal{U}$ be a non-principal ultrafilter on $\omega$, and take the ultraproduct $M = \left(\prod_{n\in \omega} M_n\right){/}\mathcal{U}$. Since both the class of models of $T$ and the class of non-models of $T'$ are elementary, they are closed under ultraproducts, so $M\models T$ and $M\not\models T'$.
On the other hand, the answer is yes in general, and we can even find examples where one of $T$ or $T'$ is finitely axiomatizable. Let $T'$ be the empty theory (which is finitely axiomatizable!). Let $\varphi_n$ be the sentence asserting that there are at most $n$ elements. Let $T$ consistent of the sentences $\varphi_n\to P$ for all $n\in \omega$.
Then $T$ and $T'$ agree on infinite models. Every infinite structure is a model of both $T$ and $T'$, since infinite structures fail to satisfy all $\varphi_n$. And both $T$ and $T'$ have arbitrarily large finite models (any finite structure in which $P$ is true is a model of both). But for all $n$, letting $M_n$ be a structure of size $n$ in which $P$ is false, $M_n\models T'$ but $M_n\not\models T$, since the sentence $\varphi_{n+1}\to P$ is false in $M_n$.
- 86,811
-
Okay, so $T$ and $T'$ can't both be finitely axiomatizable, but can just one of $T$ or $T'$ be finitely axiomatizable? – user107952 Feb 12 '25 at 22:47
-
1@user107952 Good question. I've replaced the examples in my answer with an even simpler example in which $T'$ (but not $T$) is finitely axiomatizable. – Alex Kruckman Feb 13 '25 at 01:59