差 2 也算赢
2 Below I
题目详情
你和朋友各自选择 1..100 的整数。若你选的数严格更大,则你赢 1;若相等则 0;但如果你选的数恰好比对方小 2,你也算赢。
双方都最优博弈时,最优策略是一个混合策略:从某分布抽随机变量 作为选择。
求 。
You and friend play a game where you both select an integer . The winner receives $1 from the loser. The winner is the player who selects the strictly higher number. If there is a tie, then nothing happens. However, a player can also win by selecting a value exactly below the larger integer. For example, if you select and your friend selects , you are the winner in this case. Assume both you and your friend play optimally. The optimal strategy here is a mixed strategy, where you select a random value from some appropriately determined distribution. Find .
解析
注意到任意 都被 支配(因为 额外还能赢 ),因此均衡只会在 98、99、100 上混合。
为避免被对方针对,必须在 98、99、100 上取均匀分布。
于是 ,
Original Explanation
Suppose you were planning to select a value . Note that is strictly better than , as will beat all integers would beat, as well as the integer . Therefore, whenever you can select , you should select . This means that your strategy should be to select or with some probabilities.
By the symmetry of the game, your friend should also select those values with the same probabilities. In particular, if you select with equal probability, no matter what probabilities your friend selects and with, the expected payout for each player is by symmetry. This is because we see that is beat by , is beat by , and is beat by , so each of the three outcomes is dominated by one other outcome.
Furthermore, if you select a non-uniform distribution on for your values, there exists a strategy your friend can select that yields positive expected payout for them. The probabilities of selecting each of the values for you would be and . For a numerical demonstration, say and . Then your friend should always select the value that maximizes their probability of winning. In this case, they should select , as the expected payout would be . Namely, if is the value assigned to probability , then your friend should always select the value that beats . Therefore, to eliminate this opportunity, you should select a uniform distribution among .
This means , so and .