[guided]The idea is to iterate the edge-Ramsey strategy one dimension up. In the edge case, we picked a vertex $x_i$ and, using only two possible edge colours, pigeonholed on the colours of the edges from $x_i$ into the remaining infinite set. Here we cannot pigeonhole directly because the colours of $(r-1)$-subsets through $x_i$ are parameterised by an entire family $F$, not by single elements. The fix is to use the inductive hypothesis in dimension $r - 1$ instead of pigeonhole.
Concretely, we build by recursion on $i$:
- elements $x_1 < x_2 < \cdots$ of $\mathbb{N}$,
- a decreasing chain $\mathbb{N} \supsetneq S_1 \supsetneq S_2 \supsetneq \cdots$ of infinite sets with $x_i \in S_{i-1} \setminus S_i$,
- colours $c_i \in \{\text{red}, \text{blue}\}$,
such that **every $(r-1)$-subset $F$ of $S_i$ satisfies $c(\{x_i\} \cup F) = c_i$**. This is the property (\*) on which the whole argument hinges.
*Why this is exactly what we need.* Later, for any indices $i_1 < \cdots < i_r$, we want $c(\{x_{i_1}, \ldots, x_{i_r}\})$ to depend only on $i_1$. Property (\*) applied at index $i_1$ says that $c(\{x_{i_1}\} \cup F) = c_{i_1}$ for every $F \in S_{i_1}^{(r-1)}$. If we can guarantee $\{x_{i_2}, \ldots, x_{i_r}\} \in S_{i_1}^{(r-1)}$, we are done. Since $i_2, \ldots, i_r > i_1$ and the $S_j$ are nested, $x_{i_2}, \ldots, x_{i_r} \in S_{i_1}$ provided each later $x_j$ is chosen inside $S_{j-1}$ (which we arrange by taking $x_{j} \in S_{j-1}$).
*Base step ($i = 1$).* Fix $x_1 := 1$. To find $S_1$, we look at $(r-1)$-subsets through $x_1$, i.e. we consider the induced colouring
\begin{align*}
c_{x_1}: (\mathbb{N} \setminus \{x_1\})^{(r-1)} &\to \{\text{red}, \text{blue}\} \\
F &\mapsto c(\{x_1\} \cup F).
\end{align*}
This is a 2-colouring of the $(r-1)$-subsets of the infinite set $\mathbb{N} \setminus \{x_1\}$. The inductive hypothesis at level $r - 1$ requires a 2-colouring of $\mathbb{N}^{(r-1)}$, but the statement applies equally to any infinite set (relabelling $\mathbb{N} \setminus \{x_1\}$ order-isomorphically to $\mathbb{N}$ transports the colouring). Applying it, we obtain an infinite $S_1 \subseteq \mathbb{N} \setminus \{x_1\}$ on which $c_{x_1}$ is constant, with value $c_1$. Property (\*) at $i = 1$ is immediate: for $F \in S_1^{(r-1)}$, $c(\{x_1\} \cup F) = c_{x_1}(F) = c_1$.
*Inductive step.* Having built $x_1, \ldots, x_i$, $S_1 \supsetneq \cdots \supsetneq S_i$, and $c_1, \ldots, c_i$, we choose $x_{i+1} := \min S_i$. Since $x_i \notin S_i$ (by construction), $x_{i+1} > x_i$. The induced colouring
\begin{align*}
c_{x_{i+1}}: (S_i \setminus \{x_{i+1}\})^{(r-1)} &\to \{\text{red}, \text{blue}\} \\
F &\mapsto c(\{x_{i+1}\} \cup F)
\end{align*}
is a 2-colouring of the $(r-1)$-subsets of the infinite set $S_i \setminus \{x_{i+1}\}$. By induction, we obtain an infinite $S_{i+1} \subseteq S_i \setminus \{x_{i+1}\}$ on which $c_{x_{i+1}}$ is constant with value $c_{i+1}$. Note that $S_{i+1} \subsetneq S_i$ (strictly, since $x_{i+1}$ is excluded), and $x_{i+1} \notin S_{i+1}$, so the strict-containment invariant is preserved. Property (\*) at $i+1$ follows.[/guided]