7
$\begingroup$

Consider the following game:

Two real numbers are independently, randomly, uniformly chosen from the interval [0, 1].

Andrew is secretly shown the two numbers. He must choose one of the numbers and reveal it to his opponent, Bndrew. Bndrew then guesses whether the revealed number was the larger of the two. If he gets it right, he wins; otherwise, Andrew wins.

Assuming optimal play, how likely is Bndrew to win?

$\endgroup$
1
  • $\begingroup$ Agree - it's not made clear why the answer might be believed to be anything other than 50%, or in any way different from choosing (and guessing) "heads or tails". Now, if the range were numbers from 0-infinity, then the discussion could get more interesting, as it could be argued that the best strategy is always to guess "the other number is bigger", even though again it's really a 50% chance, or a 0% of Bndrew knows, or 100% if Andrew knows Bndrew knows, or... :) $\endgroup$ Commented Jul 3 at 22:16

2 Answers 2

13
$\begingroup$

I claim that Bndrew wins

50%

of the time.

Bndrew can guarantee winning with at least that probability with the strategy of

flip a coin to choose

Andrew can guarantee Bndrew wins with at most that probability with the strategy of

Reveal whichever number is closer to $\frac{1}{2}$.

For example, if Andrew reveals $\frac{1}{5}$ or $\frac{4}{5}$, Bndrew knows only that the other number is uniformly distributed over $[0,\frac{1}{5}] \cup [\frac{4}{5},1]$ and wins half the time.

$\endgroup$
2
  • 1
    $\begingroup$ Surely Andrew could also follow a simpler strategy, similar to Bndrew’s. $\endgroup$
    – Sneftel
    Commented Jul 3 at 17:03
  • 4
    $\begingroup$ @Sneftel: No, that doesn't work. If Andrew just picks a random number, Bndrew can achieve a 75% win rate by guessing that any number < 0.5 is the smaller number and any number >= 0.5 is the bigger number. $\endgroup$ Commented Jul 4 at 0:15
4
$\begingroup$

Assuming optimal play, I think Bndrew cannot win more than

50% of the time.

Why?

Let $a$ and $b$ the two secret numbers such that $a \leq b$, and $x$ the number given by Andrew ($x=a$ or $x=b$).

If we exclude strategies that involve saying "largest" or "smallest" at random (which cannot exceed 50% win rate), Bndrew's choice can only rely on the value of $x$. That means any strategy will consist of choosing a set $S \subseteq [0,1]$ such that Andrew will say "largest" if $x \in S$, and say "smallest" otherwise.

What's the best possible choice for $S$? It would not make sense for $S$ to not be an interval with an upper bound lower than $1$, because it would mean that there exists $a \leq b$ such that Bndrew will say "highest" for $a$ but "lowest" for $b$. Hence, we can assume that $S=[s, 1]$ for some $s \in [0,1]$.

Now, what's the best possible choice for $s$? Well, to ensure a win no matter the choice of Andrew, we would like to maximise the probability of having $a \leq s \leq b$. This happens when $s=\frac{1}{2}$.

Therefore, the best strategy for Bndrew is to say "largest" if $x \geq \frac{1}{2}$ and "smallest" if $x < \frac{1}{2}$. There is a $\frac{1}{2}$ probability of having $a \leq \frac{1}{2} \leq b$ which ensures a win for Bndrew. If $a \leq b \leq \frac{1}{2}$, Andrew can prevent Bndrew from winning by giving $x=b$ (which happens with probability $\frac{1}{4}$) and if $\frac{1}{2} \leq a \leq b$, Andrew can prevent Bndrew from winning by giving $x=a$ (which also happens with probability $\frac{1}{4}$).

$\endgroup$
3
  • 2
    $\begingroup$ "If we exclude strategies that involve saying "largest" or "smallest" at random (which cannot exceed 50% win rate)" - why not? A mixed strategy could be better than a pure strategy. $\endgroup$
    – Deusovi
    Commented Jul 3 at 13:34
  • $\begingroup$ You're right. I think I took the problem the wrong way by trying to figure out Bndrew's best strategy and not Andrew's best strategy (which I end up doing in the last paragraph). @ralphmerridew just posted an answer that is less confused and way more concise; I encourage upvoting his answer :) $\endgroup$
    – Jujustum
    Commented Jul 3 at 14:07
  • $\begingroup$ @Deusovi : Sure, a mixed strategy can be better, but is it always implicit that mixed strategies are allowed? Having access to a random coin that is independent of the given numbers seems like a nontrivial advantage. $\endgroup$
    – Plop
    Commented Jul 4 at 8:02

Not the answer you're looking for? Browse other questions tagged or ask your own question.