jbriggs444
Science Advisor
Homework Helper
2024 Award
- 13,370
- 8,042
fresh_42 said:81. Bart and Lisa are sitting in front of a huge heap of skittles. Since both of them want to eat as many as possible, they decide to play a game. Bart has to write two (different) numbers (positive integers) on two pieces of paper. Then Lisa turns around one of the papers and guesses, whether the other number is higher or lower. If she is right, then she will get 10 skittles, otherwise Bart will get them.
Is there a strategy for Lisa which improve her chances in comparison to a 50:50 guess? And can Bart counter this strategy?
D95
Suppose that Bart has chosen numbers i and j. One is chosen at random. 0.5 probability for i, 0.5 probability for j. Without loss of generality, assume i > j.
Lisa decides on a strictly monotone increasing function f from the positive integers to the reals between 0 and 1. After seeing the chosen number, she uses this function to determine her probability of staying with the number that has been revealed.
Lisa's probability of seeing i and then winning is ##0.5 \times f(i)##
Lisa's probability of seeing j and then winning is ##0.5 \times (1 - f(j))##
Total ##0.5 + 0.5\times (f(i) - f(j))##
Since f is monotone increasing and i > j, f(i) > f(j) and Lisa's winning probability exceeds 0.5
Bart can counter this strategy by choosing very large numbers. Lisa's function is bounded and monotone increasing. It must therefore have a least upper bound that is approached arbitrarily closely for large n. (i.e. ##\lim_{n \to \infty}f(n) = U##. It follow that for any epsilon, Bart can choose a delta large enough that that f(i)-f(j) < epsilon for all i, j > delta.
Edit: You'd only asked about the existence of a counter-strategy for Bart rather than for an actual implementation. The following mixed strategy should perform adequately.
Suppose that Bart wishes to obtain a winning probability within epsilon of 0.5. Bart arbitrarily picks a positive integer n that is larger than 1/epsilon. He then picks uniformly and at random an integer i between 1 and n. He writes i on one piece of paper and writes i+1 on the other.
Lisa's chance of winning will depend on her choice of f, of course. But averaged over all the n possible picks by Bart, her winning probability exceeds 0.5 by a total of:
$$\frac{1}{n}\sum_{i=1}^n 0.5 \times(f(i+1)-f(i))$$
Since f(1) and f(n+1) are bounded by 0 below and 1 above, this sum is no more than 0.5/n. This is less than epsilon.
Lisa decides on a strictly monotone increasing function f from the positive integers to the reals between 0 and 1. After seeing the chosen number, she uses this function to determine her probability of staying with the number that has been revealed.
Lisa's probability of seeing i and then winning is ##0.5 \times f(i)##
Lisa's probability of seeing j and then winning is ##0.5 \times (1 - f(j))##
Total ##0.5 + 0.5\times (f(i) - f(j))##
Since f is monotone increasing and i > j, f(i) > f(j) and Lisa's winning probability exceeds 0.5
Bart can counter this strategy by choosing very large numbers. Lisa's function is bounded and monotone increasing. It must therefore have a least upper bound that is approached arbitrarily closely for large n. (i.e. ##\lim_{n \to \infty}f(n) = U##. It follow that for any epsilon, Bart can choose a delta large enough that that f(i)-f(j) < epsilon for all i, j > delta.
Edit: You'd only asked about the existence of a counter-strategy for Bart rather than for an actual implementation. The following mixed strategy should perform adequately.
Suppose that Bart wishes to obtain a winning probability within epsilon of 0.5. Bart arbitrarily picks a positive integer n that is larger than 1/epsilon. He then picks uniformly and at random an integer i between 1 and n. He writes i on one piece of paper and writes i+1 on the other.
Lisa's chance of winning will depend on her choice of f, of course. But averaged over all the n possible picks by Bart, her winning probability exceeds 0.5 by a total of:
$$\frac{1}{n}\sum_{i=1}^n 0.5 \times(f(i+1)-f(i))$$
Since f(1) and f(n+1) are bounded by 0 below and 1 above, this sum is no more than 0.5/n. This is less than epsilon.
Last edited:
