I came across the following problem and I'm stumped.
Players X and Y play the following game.
For $n\geq 2$, they consider a monic polynomial with degree $2n$, with undetermined coefficients (except the constant one): $X^{2n}+\square X^{2n-1}+\ldots+\square X+1$
The players take turn to fill the blanks with a real number of their choice.
X wins, if, when completing the last blank square available, the resulting polynomial has no real root. Otherwise Y wins.
If X plays first, who has a winning strategy ?
$Y$ always wins. The strategy for $Y$ is: play all even coefficients as long as there are any; the numbers chosen don't matter until the last move. At $Y$'s final move, play an even coefficient if there is one, otherwise play the coefficient with the smaller exponent; it suffices to play a sufficiently large negative number.
This proof is inspired by Ross Millikan's post. Let us play as $Y$. We only play even coefficients as long as there are any left to play. Since the game starts out with $n-1$ even coefficients and $n$ odd coefficients, it follows that at $Y$'s final move, there will be at most one even coefficient left. If there is an even coefficient, then $Y$ will choose to play it, and we win by choosing a sufficiently large negative number, by the reasoning in Ross's last paragraph. Otherwise there remain two odd coefficients corresponding to $x^{d_1}$ and $x^{d_2}$ for some odd $d_1<d_2$. The polynomial is now of the form $$p(x)+ax^{d_1}+bx^{d_2}$$ for some polynomial $p(x)$. Choose a small $\epsilon>0$ ($\epsilon=1/2$ will suffice). Let $$\delta=\min(\frac12, \dfrac12 \dfrac{1-2\epsilon^{d_2-d_1}}{2\epsilon^{d_2}}),$$ and choose $N>0$ such that $|p(x)/N|\leq \delta$ for all $x\in[-1,1]$. Player $Y$ will now choose to play $a=-N/\epsilon^{d_1}$. Then $p(x)+ax^{d_1}<(-1+\delta)N<0$ at $x=\epsilon$ and thus $p(x)+ax^{d_1}$ has a real root. To compensate, $X$ must play as his final move some $b\geq (1-\delta)N/\epsilon^{d_2} \geq N/(2\epsilon^{d_2})$ in order to make $p(x)+ax^{d_1}+bx^{d_2}$ positive at $x=\epsilon$. But now, at $x=-1$ we get $$p(1)-a-b<(\delta+\dfrac1{\epsilon^{d_1}}-\dfrac1{2\epsilon^{d_2}}) N = (\delta-\dfrac{1-2\epsilon^{d_2-d_1}}{2\epsilon^{d_2}})N<-\delta N<0$$ and thus $p(x)+ax^{d_1}+bx^{d_2}$ has a real root. Player $Y$ wins!