Steven Clontz |
---|
University of South Alabama |
Many topological properties may be characterized in terms of who can guarantee victory (and how much memory they require) in certain two-player, infinite-length games. To illustrate this, we will explore a novel game-theoretic proof of the uncountability of the real numbers, and then discuss the class of "selection games" used throughout general topology.
Let \(W\subseteq\mathbb R\text{.}\) Then \(CG(W)\) denotes the following game played by players \(P1\) and \(P2\text{.}\)
During round \(0\text{,}\) \(P1\) chooses some real number \(a_0\text{,}\) followed by \(P2\) choosing \(b_0\) such that \(a_0\lt b_0\text{.}\)
During round \(n+1\text{,}\) \(P1\) chooses \(a_{n+1}\in(a_n,b_n)\text{,}\) followed by \(P2\) choosing \(b_{n+1}\in(a_{n+1},b_n)\text{.}\)
After \(\omega=\{0,1,2,\dots\}\)-many rounds, \(P1\) wins this game in the case that \(\displaystyle\lim_{n\to\infty}a_n\in W\text{.}\)
(This game seems to be introduced by Grossman and Turett in the problems section of Mathematics Magazine Vol. 71, with some further analysis done by Matt Baker in a later volume.)
It'd take some time to complete this kind of \(\omega\)-length game.
However, each player in such a game can define a strategy \(\sigma:M^{\lt\omega}\to M\) that designates their moves in response to their opponent's actions...
Typically, we want to prove something about the existence of winning strategies that defeat any counter-strategy chosen by the opponent.
If player \(P\) has a winning strategy for game \(G\text{,}\) we write \(P\win G\text{.}\)
\(P1\win CG(\mathbb R)\text{.}\)
If \(W\) is countable, then \(P2\win CG(W)\text{.}\)
\(\mathbb R\) is uncountable.
If \(W\) is countable, then \(P2\win CG(W)\text{.}\)
Let \(W=\{w_n:n\in\omega\}\text{.}\) Then if any element of \(W\) is legal to play during \(P2\)'s turn, their strategy is to play \(w_n\text{,}\) where \(n\) is the minimal ordinal such that \(w_n\) is legal to play. Otherwise, \(P2\) chooses arbitrarily.
At the end of the game, each element \(w_N\in W\) was either played before round \(N\text{,}\) or was illegal to be played in round \(N\text{.}\) Either way, \(w_N\not=\lim_{n\to\infty} a_n\text{,}\) so \(P2\)'s strategy is winning.
Given two sets \(\mc A,\mc B\text{,}\) the selection game \(G_1(\mc A,\mc B)\) is played as follows:
\(P2\) wins this game provided \(\{b_n:n\lt\omega\}\in\mc B\text{.}\)
\(G_{fin}(\mc A,\mc B)\) is similar, except \(P2\) can make finitely-many selections each round.
Let \(X\) be some topological space with a particular point \(x\text{.}\)
A strategy that can ignore all but the most recent move of the opponent (but might use the current round #) is said to be Markov.
A strategy that can ignore all moves of the opponent (but probably uses the current round #) is said to be predetermined.
For \(X=\mathbb R\) with the usual topology:
Notes:
Let \(\mc O\) be the set of open covers, and \(\mc P\) be the set of point-bases. Let \(\neg\) denote the complement of a set.
Then the point-open game \(G_1(\mc P,\neg\mc O)\) essentially has \(P1\) choosing points, and \(P2\) selecting a neighborhood for each point. Then \(P1\) wins if \(P2\)'s sets form a cover.
Compare with \(G_1(\mc O,\mc O)\text{:}\) \(P1\) chooses open covers, \(P2\) selects an open set from the cover. Here \(P2\) wants to build the open cover.
Galvin (1978) demonstrated the following:
The following also holds:
All together, we say that these games are dual.
It turns out that the key to this proof comes from a simple observation:
“For each \(x\in X\text{,}\) pick any neighborhood \(U_x\text{.}\) Then \(\{U_x:x\in X\}\) is an open cover.”
Put another way:
“For each \(B_x\in\mc P\text{,}\) pick any \(U_x\in B_x\text{.}\) Then \(\{U_x:B_x\in\mc P\}\in\mc O\text{.}\)”
So let's say \(\mc R\) reflects* \(\mc A\) given the following:
“For each \(R\in\mc R\text{,}\) pick any \(b_R\in R\text{.}\) Then \(\{b_R:R\in \mc R\}\in\mc A\text{.}\)” (*slight simplification)
So \(\mc P\) refelcts \(\mc O\text{,}\) and \(G_1(\mc O,\mc O)\) and \(G_1(\mc P,\neg\mc O)\) are dual.
Likewise, let \(\mc T\) be the topology of a space, and choose a point from each open set. That collection is a dense subspace, a member of \(\mc D\text{.}\)
What can we say about \(G_1(\mc D,\Gamma_x)\) and \(G_1(\mc T,\neg\Gamma_x)\text{?}\)
YES, they are dual!
If \(\mc R\) reflects \(\mc A\text{,}\) then \(G_1(\mc A,\mc B)\) and \(G_1(\mc R,\neg\mc B)\) are dual.