The relevant definitions regarding notation, and the statements of Ramsey's theorems, can be found at the beginning of this question. There, I asked, and later provided an answer (mostly correct, as far as I can tell) to the question of how to prove the finite Ramsey theorem ($\mathrm{FRT}$) from the infinite Ramsey theorem ($\mathrm{IRT}$) via propositional compactness.

When I can manage to do so, and as I am still a novice in the subject, I find it quite helpful trying to prove whatever result I know how to prove using propositional compactness via first-order compactness as well.

The idea is to use first-order compactness to show that if $\mathrm{FRT}$ fails for some $k_{0}\in\mathbb{N}^{+}$ then $\mathrm{IRT}$ also fails. Therefore, we need to define appropriate first-order language $L$ and set of $L$-sentences $S_{k_{0}}$ encoding $\neg\mathrm{IRT}$ for $k_{0}$. Then we will prove finite satisfiability of $S_{k_{0}}$ assuming $\neg\mathrm{FRT}$ for $k_{0}$ and apply first-orer compactness. This will yield an model in which $\mathrm{IRT}$ fails.

Here is is my attempt at defining $L$ and $S_{k_{0}}$ of $L$-sentences.

The language $L$ contains:

  1. $p$-ary predicate symbols $C^{1}, C^{2},\ldots,C^{r}$ (intuitively $C^{i}x_{1}\ldots x_{p}$ means $c(\{x_{1},\ldots,x_{p}\})=i$ when true).
  2. Equality $=$.

The set $S_{k_{0}}$ consists of the following $L$-sentences:

  1. $\forall x_{1}\ldots\forall x_{p}\,(C^{i}x_{1}\ldots x_{p}\Leftrightarrow C^{i}x_{f(1)}\ldots x_{f(p)})$ for all $1\leq i\leq r$ and all permutations $f$ of $[p]$.
  2. $\displaystyle\forall x_{1}\ldots\forall x_{p}\,\left(C^{i}x_{1}\ldots x_{p}\Rightarrow\left(\bigwedge_{1\leq u<v\leq p}\neg(x_{u}=x_{v})\right)\right)$ for all $1\leq i\leq r$.

1 and 2 mean that only the $p$-element subsets of our domain may be colored.

  1. $\forall x_{1}\ldots\forall x_{p}\,\left(\displaystyle\left(\bigwedge_{1\leq u<v\leq p}\neg(x_{u}=x_{v})\right)\Rightarrow\bigvee_{1\leq i\leq r}C^{i}x_{1}\ldots x_{p}\right)$ (each $p$-element subset is assigned at least one color).
  2. $\forall x_{1}\ldots\forall x_{p}\,\neg(C^{i}x_{1}\ldots x_{p}\wedge C^{j}x_{1}\ldots x_{p})$ for all $1\leq i<j\leq r$ (each $p$-element subset is assigned at most one color).
  3. $\displaystyle\forall x_{1}\ldots\forall x_{k_{0}}\,\left(\left(\bigwedge_{1\leq u<v\leq k_{0}}\neg(x_{u}=x_{v})\right)\Rightarrow\left(\bigvee_{1\leq i_{1}<\cdots<i_{p}\leq k_{0}}\neg C^{i}x_{i_{1}}\ldots x_{i_{p}}\right)\right)$ for all $1\leq i\leq r$ (every $k_{0}$-element subset of our domain has at least one $p$-element subset which is not colored by $i$).

5 means that there is no homogeneous $k_{0}$-element subset for the $r$-coloring of the set of $p$-element subsets of our domain.

  1. $\displaystyle\exists x_{1}\ldots\exists x_{n}\,\left(\bigwedge_{1\leq u<v\leq p}\neg(x_{u}=x_{v})\right)$ for all $n\in\mathbb{N}$ (Our domain has infinitely many elements).
  2. The equality $=$ axioms.

Question: Did I define $L$ and $S_{k_{0}}$ correctly? I am having a bit of trouble showing the finite satisfiability of $S_{k_{0}}$

  • $\begingroup$ Good question, and good work already! Note 8 follows from 1 (and what are you using these constants for..?). I think it might make your life easier if you stick to the case of just $2$ colours (which is enough to prove full Ramsey without too much trouble!). The way you'll prove finite satisfiability is to note that a finite fragment of your theory will only be able to assert "I have at least $N$ elements", for some $N$, so you can use $\lnot$ FRT to take a bad colouring of $[N]$, which should model your theory! Can you explain more what was giving you trouble? $\endgroup$ Apr 14, 2023 at 19:39
  • 1
    $\begingroup$ FWIW, the use of constants isn't necessarily silly here! EG, if you want to do an argument like this about van der Waerden's theorem, they'll come in handy. But I think using the compactness theorem from logic is a bit unwieldy here anyway - there is a notion of "compactness argument" in Ramsey theory which is much easier to generalise to more complicated situations, which works by stitching together an infinite sequence of bad colourings. See eg the proof of Theorem 3 here. It's a good trick that's worth getting used to! $\endgroup$ Apr 17, 2023 at 17:17
  • $\begingroup$ Or an argument like this about graph colorings. Constants will come in handy there as well. In any case, it was good practice of the basics, which is what I need. Thank you for the link. Good tip! $\endgroup$
    – John
    Apr 17, 2023 at 18:35
  • $\begingroup$ Do we really need a compacness argument to prove that $IRT\Rightarrow FRT$? If $FRT$ fails for $k,p,r$, then for each $n\in\mathbb{n}$ there exists an $r$-coloring $c_{n}$ of $[n]^{p}$ with no homogeneous subset of size $k$. Define $c:[\mathbb{N}]^{p}\to[r]$ by $c(A) = c_{b}(A-\{b\})$, where $b=\max\{a:a\in A\}$. By $IRT$ the $r$-coloring $c$ has an infinite homogeneou subset $H\subseteq\mathbb{N}$. But then, for any size $k$ subset $H_{k}\subset H$ there is some $b\in H$ such that $c|_{b}[H_{k}]^{p}$ is constant. That is, $H_{k}$ is a homogeneous subset of size $k$ for $c_{b}$. Contradiction! $\endgroup$
    – John
    Apr 17, 2023 at 18:50
  • $\begingroup$ I'm not sure I follow! What does $c_b(A - \{b\})$ mean if $c_b$ is defined on $p$-sets? If you meant $c(A) = c_b(A)$, then a set $H$ being homogeneous for $c$ doesn't necessarily imply any subset $H_k$ is homogeneous for some $c_n$, since the $p$-subsets of $H_k$ will be coloured from different $c_n$, right? For example, say $c_n(A)$ colours $A$ red if $n \in A$, or otherwise blue if $1 \in A$, or else green. Then $H = \Bbb N$ is homogeneous for $c$, but $\{1, ..., k\}$ is not homogeneous for any $c_n$ (if $k > p$). So I think there has to be some cleverness! $\endgroup$ Apr 17, 2023 at 20:02

1 Answer 1


In order to remove the question from the unanswered list, I will provide a proof that $S_{k_{0}}$ is finitely satisfiable assuming $\neg\mathrm{FRT}$ for $k_{0}$.

Indeed, let $\Delta\subset S_{k_{0}}$ be finite. This means there is a largest $m\in\mathbb{N}$ for which a sentence of the from 6 occurs in $\Delta$. Therefore, $\Delta$ says "I will have at least $m$ elements on any model". That is, the domain of any model $M$ satifying $\Delta$ will have at least $m$-elements. Let $N>m$, and recall that since $\neg\mathrm{FRT}$ for $k_{0}$, there is an $r$-coloring $c_{N}:[N]^{p}\rightarrow [r]$ having no homoegeous subset of size $k_{0}$. We define a model $M$ as follows:

  1. $U_{M}=[N]$.
  2. $C^{i}_{M} = \{(a_{1},\ldots,a_{p})\in U_{M}^{p}: \{a_{1},\ldots,a_{p}\}\in[N]^{p}\text{ and }c_{N}(\{a_{1},\ldots,a_{p}\})=i\}$ for $1\leq i\leq r$ (where $U_{M}^{p}$ denotes the cartesian product of $p$ copies of $U_{M}$).
  3. $=_{M}$, i.e. the obvious interpretation of $=$.

Now, let us check that $M=(U_{M},C^{i}_{M}:1\leq i\leq r,=_{M})$ satisfies $\Delta$:

1. $M$ satisfies the sentences of 1 in $\Delta$ becasue $C^{i}a_{1},\ldots,a_{p}$ true iff $\{a_{f(1)},\ldots,a_{f(p)}\}\in[N]^{p}$ for any permutation $f$ of $[p]$ iff $(a_{f(1)},\ldots,a_{f(p)})\in C^{i}_{M}$ for any permutation $f$ of $[p]$ iff $C^{i}a_{f(1)}\ldots a_{f(p)}$ is true.
2. $M$ satisfies the sentences of 2 in $\Delta$, since $C^{i}a_{1}\ldots a_{p}$ true iff $(a_{1},\ldots,a_{p})\in C^{i}_{M}$ iff $\{a_{1},\ldots,a_{p}\}\in[N]^{p}$, which clearly implies the $a_{j}$ are all distinct.
3. $M$ satisfies the sentences of 3 in $\Delta$ because if $\displaystyle\bigwedge_{1\leq u<v\leq p}\neg(a_{u}=a_{v})$ is true, then $\{a_{1},\ldots,a_{p}\}\in[N]^{p}$, which implies that $c_{N}$ assigns a color $i$ to $\{a_{1},\ldots,a_{p}\}$. That is, $(a_{1},\ldots,a_{p})\in C^{i}_{M}$.
4. $M$ satisfies the sentences of 4 in $\Delta$ because $c_{N}$ is a function. Therefore for $1\leq i<j\leq r$, $\neg(C^{i}a_{1}\ldots a_{p}\wedge C^{j}a_{1}\ldots a_{p})$ is true iff for $1\leq i<j\leq r$, $(a_{1},\ldots,a_{p})\not\in C^{i}_{M}\cap C^{j}_{M}$.
5. $M$ satisfies the sentences of 5 in $\Delta$ because $c_{N}$ does not have a homogeneous subset of size $k_{0}$. Therefore if $\displaystyle\bigwedge_{1\leq u<v\leq k_{0}}\neg(a_{u}=a_{v})$ is true, the set $\{a_{1},\ldots,a_{k_{0}}\}$ is not homogeneous for $c_{N}$, which implies that it is colored by at least two colors, which implies that $[\{a_{1},\ldots, a_{k_{0}}\}]^{p}$ it is colored by at least two colors, which implies that for each color $1\leq i\leq r$ there is at least one $p$-element subset of $\{a_{1},\ldots,a_{k_{0}}\}$ which is not colored by $i$; i.e. it imples that $\displaystyle\bigvee_{1\leq i_{1}<\cdots<i_{p}\leq k_{0}}\neg C^{i}a_{i_{1}}\ldots a_{i_{p}}$ is true.
6. Since $N>m$, the domain $U_{M}$ certainly has at least $m$ elements. That is, $\varphi_{n}$ is satisfied for all $0\leq n\leq m$. Therefore, $M$ satisfies the sentences of 6 in $\Delta$.
7. $M$ clearly satisfies the sentences of 7 in $\Delta$.

By first-order compactness, the set $S_{k_{0}}$ has a model $M'$ whose domain $U_{M'}$ is countably infinite. Moreover, since $M'$ satisfies $S_{k_{0}}$, any $r$-coloring $c$ of $[U_{M'}]^{p}$ fails to have a homogeneous subset of size $k_{0}$. That is, $\mathrm{IRT}$ fails for $M'$, as desired (recall that if $\mathrm{IRT}$ holds for $M'$, $c$ would have an infinite homogeneous subset, and therefore homogeneous subsets of size $k$ for all $k\in\mathbb{N}$).


You must log in to answer this question.

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