1
$\begingroup$

An uncountably categorical theory always has a strongly minimal set definable over its prime model, but sometimes this set needs parameters to define.

By a $\varnothing$-definable imaginary I mean the quotient of some finite power of the home sort by a $\varnothing$-definable equivalence relation. No added generality is gained by considering quotients of definable sets, since we can always just extend the equivalence relation and send everything in the complement to a single point.

Every example of an uncountably categorical theory I can think of has a $\varnothing$-definable strongly minimal imaginary. So the question is:

Does every uncountably categorical theory have a $\varnothing$-definable strongly minimal imaginary?

The best I've been able to come up with is the theory of $\mathbb{Z}$ with a symmetric successor relation and a predicate for $x\equiv y\text{ (mod 2})$. By a direct argument you can show that no $\varnothing$-definable equivalence relation on elements gives a strongly minimal imaginary, but I believe there is a strongly minimal quotient of 2-tuples, specifically unordered adjacent pairs.

A few easier questions that would be relevant would be the same but allowing $\text{acl}^\text{eq}(\varnothing)$-definable imaginaries or only requiring that the imaginary have Morley rank $1$, but allowing it to have Morley degree greater than 1.

$\endgroup$

1 Answer 1

1
$\begingroup$

I honestly thought about a lot of different possible counterexamples before realizing that a small modification of the structure I already mentioned gives a counterexample. Specifically if we consider the structure of $\mathbb{Z}$ with the successor function and a predicate $P(x,y)$ for $x\equiv y\text{ (mod 2})$. Let $T$ be the theory of this structure.

To see this assume that $T$ has a $\varnothing$-definable strongly minimal imaginary $I$. Since the theory in uncountably categorical there must be a definable finite-to-finite correspondence between $I$ and the definable set $P(x,0)$ (where $0$ is some element of the prime model). Let $\varphi(x,y,\overline{a})$ be that finite-to-finite correspondence (and $\overline{a}$ is a tuple of parameters in the home sort). There needs to be a uniform finite bound $n$ on the sizes of the fibers of the correspondence, so we can write a formula $\psi(w,\overline{z})$ that means '$\varphi(x,y,\overline{z})$ defines a finite-to-finite correspondence between $I$ and $P(x,w)$ with fibers of size at most $n$'. Then the theory $T$ says $\forall w \exists \overline{z} \psi(w,\overline{z})$, so in particular there is some set of parameters $\overline{a}$ in the prime model such that $\psi(x,y,\overline{a})$ is a finite-to-finite correspondence between $I$ and $P(x,0)$. Since all of the prime model is in $\text{dcl}(0)$, we can just let $\overline{a}$ actually be $0$, so we have a formula $\psi(x,y,0)$ that defines a finite-to-finite correspondence between $I$ and $P(x,0)$.

There needs to be a number $m$ such that for any $b\in I$, the largest and smallest element of the home sort related to $b$ by $\varphi(x,y,0)$ are no more than $m$ many successor steps away, (EDIT: There's a subtlety at this point which is that the set related can be arbitrarily large (it could always contain 0 for instance), but the point is it needs to generically contain a chunk around 0 of bounded size and some moving chunk of bounded size. You can define the needed function in terms of the moving chunk.) so we can actually define a formula $\chi(x,y,0)$ such that each element of $I$ is related to the least element it was related to by $\varphi(x,y,0)$, or $0$ if there were no elements related (this can only happen for finitely many elements of $I$), in other words $\chi(x,y,0)$ defines a function $f(0,x)$ from $I$ into the set $P(x,0)$. The image of this function must be all but finitely many elements of the set $P(x,0)$.

By symmetry, for any even number $2k$, $f(2k,x)$ must define an almost surjective map from $I$ to the set $P(x,0)$. Fix some $2k$. I claim that for all but finitely many $b \in I$, $f(0,x)=f(2k,x)$. If this weren't true, then $T$ would interpret the group $(\mathbb{Z},+)$, which it can't because it's $\omega$-stable. So we can define a new function over the imaginary parameter corresponding to the $P$-equivalence class of $0$, which I'll call $[0]_P$. The function, $g([0]_P,x)$, is given by the average behavior of the function $f(2k,x)$ for different values of $2k$. By stability and symmetry we only need to check some sufficiently large collection of parameters, so the definition of $g([0]_p,x)$ is something like 'there exists more than 17 elements $y$ of $[0]_P$ such that $f(y,x)$ outputs the same value for, let $g([0]_P,x)$ be that value'.

By symmetry again everything we said about the even numbers works for the odd numbers as well and we have a definable function $g([1]_P,x)$ from $I$ into the set $P(x,1)$. Since $[1]_P$ is obtained from $[0]_P$ by shifting everything up by $1$, it must be the case that $g([1]_P,x)=S(g([0]_P,x))$, but by symmetry we also have $g([0]_P,x)=S(g([1]_P,x))$, which is impossible. Therefore no such imaginary can exist.

There's still the matter of the followup question regarding $\varnothing$-definable imaginaries with finite Morley rank, but that's not the title of this question so I'm posting this as an answer rather than as an edit to the question.

$\endgroup$
1
  • $\begingroup$ Very nice example! $\endgroup$ Feb 15, 2019 at 15:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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