Suppose I play the following game against the Opponent. My moves are rational numbers $p_i$ and the Opponent's moves are real numbers $\epsilon_i>0$.

On turn $n+1$ the past move sequence is $p_1,\epsilon_1,\ldots, p_{n}, \epsilon_{n}$. I select a point $p_{n+1}\in \mathbb Q \cap (p_{n}-\epsilon_{n},p_{n}+\epsilon_{n})$ and the Opponent selects some $\epsilon_{n+1} >0$.

I win provided the sequence $p_i$ tends to an irrational number.

Does anyone know if I have a winning strategy? Certainly I don't if the rational and irrational numbers are swapped, as the Opponent can just enumerate the rationals and select the intervals small enough to exclude each rational in turn.

I imagine I can always win. But that's only because this looks like the Banach-Mazur game, which I can win if the target set is the irrationals.

The motivation behind this is I want to recursively build a set of homeomorphisms $F_1,F_2,\ldots: \mathbb R \to \mathbb R$ such that $F(x) =\displaystyle \lim_{n \to \infty} F_n \circ F_{n-1} \circ \ldots \circ F_1 (x)$ is a well-defined homeomorphism and such that some fixed $p \in \mathbb Q$ is sent to an irrational number.

I have conditions under which the limit exists. Namely I have to ensure the next $\max d(x,F_n(x)) < \epsilon_n$. Unfortunately $\epsilon_n$ are not known in advance. Each $\epsilon_n$ is determined by $F_1,\ldots, F_{n-1}$. This leads to the above game where $p_n = F_n \circ F_{n-1} \circ \ldots F_1(p)$.


Sure you can win. Let's enumerate rationals as $q_n, n = 1, 2,\ldots$. Also we can WLOG assume that $\varepsilon_{i+1} \le \frac{\varepsilon_i}{20000}$. We will make it so that $d(q_n, p_{n+1}) \ge 10\varepsilon_{n+1}$. Then for $m > n$ we have $d(q_n, p_m) \ge 10\varepsilon_{n+1} - \varepsilon_{n + 2} - \ldots \ge \varepsilon_{n+1}$.

For $p_{n+1}$ the possible values are an interval of length $\varepsilon_n \ge 20000\varepsilon_{n+1}$. Therefore there exists at least one admissible value of $p_{n+1}$ such that $d(q_n, p_{n+1})\ge 10\varepsilon_{n+1}$ whatever the value of $q_n$ is.

Thus, $p_n$ can not converge to any rational $q$ since if $q = q_n$ for some $n$ then $d(q, p_m) \ge \varepsilon_{n+1} > 0$ for $m > n$. On the other hand since $\varepsilon_{i+1} \le \frac{\varepsilon_i}{20000}$ the sequence $p_n$ is Cauchy and thus converge for some $q\in \mathbb{R}$. Therefore $q\in \mathbb{R}\backslash \mathbb{Q}$.

  • $\begingroup$ Cool, thanks. It's not important to my problem, but any ideas what happens if we replace $\mathbb Q$ with some arbitrary comeagre set, or if this game can be reduced to Banach-Mazur or something more well-known? $\endgroup$ – Daron Dec 10 '19 at 10:25

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.