How do some Ramsey's Theorem proofs get to proving the inequality:
$R(s, t) \le R(s − 1, t) + R(s, t − 1)$
I get how the result shows, in the end, that there is a finite $R(s,t)$ but what's the intuition behind that step. In other words how does one think of such an inequality?
The proof of such inequality is similar. Again, we fix a point $p$ of $K_n$ whose edges are randomly and independently colored by red or blue, where $$n=R(s,t-1)+R(s-1,t).$$
We then partition the neighbors (of course all the vertices in $K_n$ other than $p$) into two sets: $A$ and $B$ such that the edges between $p$ and vertices in $A$ are colored in red and the edges between $p$ and vertices in $A$ are colored in blue. As we can see, either \begin{equation} |A|\geq R(s-1,t)\quad\text{or}\quad|B|\geq R(s,t-1). \end{equation}
If $|A|\geq R(s-1,t)$, consider the clique induced by $A$. Such clique either contains a red $K_{s-1}$, which, together with $p$, form a red $K_s$, or a blue $K_t$. Similarly, if $|B|\geq R(s,t-1)$, then the clique induced by $B$ either contains a blue $K_{t-1}$ or a red $K_s$. That is, we can either find a red $K_s$ or a blue $K_t$ in such $K_n$, hence $$R(s,t)\leq R(s,t-1)+R(s-1,t).$$