Prev Next Top Problem15_1

 \newcommand{\P}{{\bf P}}
Problem 15.1 of the Text

The Game
We consider a game where players  A and  B simultaneously choose a coin (which they own) of value  x or  y where  x \neq y . If the chosen coins have the same value, player  A gets to keep them. Otherwise player  B gets to keep them. We use  P( b , a ) to denote the payoff from player  B to player  A when  A chooses  a and  B chooses  b . The following table gives the payoff from player  B to player  A for each of the possible cases:  \[
\begin{array}{ll}
P(x, x) =  x  & P(x, y) =  -y \\
P(y, x) =  -x & P(y, y) =   y
\end{array}
\] 


Strategy for A
The optimal strategy for  A (if the strategy is discovered by  B ) solves the problem  \[
{\rm maximize} \left[ \inf_{b \in \P(2)}   
     b_1 x a_1 - b_1 y a_2 - b_2 x a_1 + b_2 y a_2
\right] \; w.r.t. \; a \in \P(2)
\] 
Applying the inf part of the lemma for matrix games, we obtain the equivalent problem:  \[
{\rm maximize} \left[ 
     \min \left( x a_1 - y a_2 \; , \; - x a_1 + y a_2 \right)
\right] \; w.r.t. \; a \in \P(2)
\] 
Because there are only two components of  a , we can use  a_2 = 1 - a_1 to replace  a_2 in the objective and obtain the equivalent problem  \[
{\rm maximize} \left[ 
     \min \left( (x + y) a_1 - y \; , \; y - (x + y) a_1 \right)
\right] \; w.r.t. \; a_1 \in [0, 1]
\] 
The function  \[
f( a_1 ) =
     \min \left( (x + y) a_1 - y \; , \; y - (x + y) a_1 \right)
\] 
has constant derivative on the interval  [0,1] except for three points. Thus, there are three possible solutions for the maximum:  a_1 = 0 ,  a_1 = 1 or the solution of the following equation  \[
\begin{array}{rcl}
(x + y) a_1 - y  & = & y - (x + y) a_1 \\
a_1              & = & y / (x + y)
\end{array}
\] 
The corresponding objective function values are  \[
\begin{array}{rclcrcl}
0 & = & a_1 & \Rightarrow & -y & = &
      \min \left( (x + y) a_1 - y \; , \; y - (x + y) a_1 \right)
\\
1 & = & a_1 & \Rightarrow & -x & = &
      \min \left( (x + y) a_1 - y \; , \; y - (x + y) a_1 \right)
\\
y / (x + y) & = & a_1 & \Rightarrow & 0 & = &
      \min \left( (x + y) a_1 - y \; , \; y - (x + y) a_1 \right)
\end{array}
\] 
It follows (from the fact that  x > 0 and  y > 0 ) that the optimal strategy for  A is to choose  x with probability  y/(x + y) and to choose  y with probability  x/(x + y) . In this case, the expected payoff of the game is zero (provided that player  B plays in an optimal fashion).

Strategy for B
The optimal strategy for  B (if the strategy is discovered by  A ) solves the problem  \[
{\rm minimize} \left[ \sup_{a \in \P(2)}   
     b_1 x a_1 - b_1 y a_2 - b_2 x a_1 + b_2 y a_2
\right] \; w.r.t. \; b \in \P(2)
\] 
Applying the sup part of the lemma for matrix games, we obtain the equivalent problem:  \[
{\rm minimize} \left[ 
     \max \left( b_1 x - b_2 x \; , \; - b_1 y + b_2 y \right)
\right] \; w.r.t. \; b \in \P(2)
\] 
Because there are only two components of  b , we can use  b_2 = 1 - b_1 to replace  b_2 in the objective and obtain the equivalent problem  \[
{\rm minimize} \left[ 
     \max \left( x ( 2 b_1 - 1 ) \; , \; (1 - 2 b_1 ) y \right)
\right] \; w.r.t. \; b_1 \in [0, 1]
\] 
The function  \[
g( b_1 ) =
     \max \left( x ( 2 b_1 - 1 ) \; , \; (1 - 2 b_1 ) y \right)
\] 
has constant derivative on the interval  [0,1] except for three points. Thus, there are three possible solutions for the maximum:  b_1 = 0 ,  b_1 = 1 or the solution of the following equation  \[
\begin{array}{rcl}
x ( 2 b_1 - 1 ) & = & ( 1 - 2 b_1 ) y
\\
2 (x + y) b_1   & = & (x + y)
\\
b_1             & = & 1 / 2
\end{array}
\] 
The corresponding objective function values are  \[
\begin{array}{rclcrcl}
0 & = & b_1 & \Rightarrow & y & = &
     \max \left( x ( 2 b_1 - 1 ) \; , \; (1 - 2 b_1 ) y \right)
\\
1 & = & b_1 & \Rightarrow & x & = &
     \max \left( x ( 2 b_1 - 1 ) \; , \; (1 - 2 b_1 ) y \right)
\\
1 / 2 &  =  & b_1 & \Rightarrow & 0 & = &
     \max \left( x ( 2 b_1 - 1 ) \; , \; (1 - 2 b_1 ) y \right)
\end{array}
\] 
It follows (from the fact that  x > 0 and  y > 0 ) that the optimal strategy for  B is to choose  x with probability  1/2 and to choose  y with probability  1/2 . In this case, the expected payoff of the game is zero (provided that player  A plays in an optimal fashion).
Input File: Problem15_1.omh