5
$\begingroup$

Let ${\cal U}$ be a countable random graph. Prove that for every formula $\varphi(x)\in L({\cal U})$, where $x$ has arbitrary finite arity, there are a positive integer $n$ and finite $C\subseteq{\cal U}^n$, such that any automorphism of ${\cal U}$ fixes $\varphi({\cal U})$ setwise iff it fixes $C$ setwise.

Motivation The claim above says that the theory of the random graph has weak elimination of imaginaries. This is a well-known fact and the proof should be folklore. But I could only come up with very messy arguments.

$\endgroup$
3
  • $\begingroup$ What does $\varphi(\mathcal U)$ mean? $\endgroup$
    – bof
    Jun 2 '16 at 10:59
  • $\begingroup$ @bof: points in $\mathcal U$ satisfying $\varphi$. $\endgroup$
    – tomasz
    Jun 2 '16 at 11:08
  • 1
    $\begingroup$ You're right with regard to the hole in my argument. But I think a simpler one should actually work, though I am not sure about the details: if you have a formula $\varphi(\bar x, \bar c)$, then we can find an equivalent formula (by "throwing away superfluous variables" from a qf equivalent formula) $\varphi'(\bar x,\bar c')$ where $\bar c'\subseteq \bar c$ such that $\varphi'(\bar x,\bar c')\equiv \varphi'(\bar x,\bar d)$ implies that $\bar d$ is a permutation of $\bar c'$. $\endgroup$
    – tomasz
    Jun 2 '16 at 12:31
1
$\begingroup$

Here is an argument using automorphisms. I'm not sure if this is quite as clean an argument as you are looking for, but hopefully it's at least a step in the right direction. Let $D\subseteq\mathcal{U}^k$ be any $\mathcal{U}$-definable set, and choose a subset $A\subset\mathcal{U}$ of minimal possible size over which $D$ is definable. Now, if $B$ is any other set over which $D$ is definable, then by Lemma 2 below $D$ is definable over $A\cap B$, so by minimality of $|A|$ we have $A\subseteq B$. In particular, any automorphism of $\mathcal{U}$ that fixes $D$ setwise must also fix $A$ setwise, so the set $A$ gives the desired parameter for $D$.


Lemma 1: Let $A,B\subset\mathcal{U}$ be finite sets and $\overline{c},\overline{d}\in\mathcal{U}^k$ a pair of $k$-tuples. If $\overline{c}\cap A=\overline{d}\cap B=\varnothing$ and $\overline{c}\equiv_{A\cap B}\overline{d}$, then there exists a $k$-tuple $\overline{e}$ with $\overline{e}\equiv_A\overline{c}$ and $\overline{e}\equiv_B\overline{d}$.

Proof: Proceeding inductively, and replacing $A$ with $A\cup\{c_1,\dots,c_{i-1}\}$ and $B$ with $B\cup\{d_1,\dots,d_{i-1}\}$ at the $i$-th step, we may assume that $k=1$ and thus replace the tuples $\overline{c}$ and $\overline{d}$ with the elements $c=c_1$ and $d=d_1$. Now, write $A=A_0\sqcup A_1$ and $B=B_0\sqcup B_1$, where $A_0$ (resp. $B_0$) is the set of all $a\in A$ (resp. $b\in B$) such that $c R a$ (resp. $d R b$). By quantifier elimination, $\operatorname{tp}(c/A)$ is isolated by $$\{v R a:a\in A_0\}\cup\{\neg(v Ra):a\in A_1\}\cup\{v\neq a:a\in A\},$$ and similarly for $\operatorname{tp}(d/B)$. Hence we need only find an element outside $A\cup B$ that is $R$-related to every element in $A_0\cup B_0$ and not $R$-related to any element in $A_1\cup B_1$, and this is possible by the defining property of the random graph. $\blacksquare$


Lemma 2: Suppose $D$ is $A$-definable and $B$-definable for finite sets $A,B\subset\mathcal{U}$. Then $D$ is $A\cap B$-definable.

Proof: Let $f$ be any automorphism fixing $A\cap B$ pointwise; we want to show that $f$ fixes $D$ setwise. To see this, enumerate $A\setminus B$ as a tuple $\bar{c}$. Since $f$ fixes $A\cap B$, we have $f(\bar{c})\cap (A\cap B)=\varnothing$, so we may find a conjugate $\bar{c}'\equiv_B f(\bar{c})$ with $\bar{c}'\cap A=\varnothing$. Let $g$ be any automorphism fixing $B$ pointwise and taking $f(\bar{c})$ to $\bar{c}'$; then $g$ fixes $D$ setwise.

Since $\bar{c}'\cap A=\varnothing$ and $\bar{c}\cap B=\varnothing$, and $\bar{c}$ and $\bar{c}'$ are conjugate by the $(A\cap B)$-automorphism $g\circ f$, by Lemma 1 we may find a tuple $\bar{c}''$ with $\bar{c}'\equiv_A\bar{c}''$ and $\bar{c}''\equiv_B \bar{c}$. Let $h$ be the composition of any two automorphisms witnessing these; since $h$ is the composition of an automorphism fixing $A$ pointwise and one fixing $B$ pointwise, we have that $h$ fixes $A\cap B$ pointwise and fixes $D$ setwise. On the other hand, $h$ takes $\bar{c}'=g(f(\bar{c}))$ to $\bar{c}$, whence $h\circ g\circ f$ fixes $A\setminus B$ pointwise. So $h\circ g\circ f$ fixes $A$ pointwise, and hence fixes $D$ setwise. Since $h\circ g$ fixes $D$ setwise as well, this means that $f$ does too, as desired. $\blacksquare$

$\endgroup$
2
  • 1
    $\begingroup$ Thank you. The argument seems very clean. In Lemma 1, wouldn't you need to assume $c\equiv_{A\cap B}d$? $\endgroup$ Jul 23 at 12:42
  • $\begingroup$ @PrimoPetri oh thank you, you are of course absolutely right!! have fixed :) $\endgroup$ Jul 23 at 14:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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