Professional Documents
Culture Documents
9 OD - Game Theory - H PDF
9 OD - Game Theory - H PDF
Mathematical theory that deals, in an formal, abstract
GAME THEORY way, with the general features of competitive
situations :
Like parlor games, military battles, political campaigns,
advertising and marketing campaigns etc
advertising and marketing campaigns, etc.
(Hillier & Lieberman Introduction to Operations Research, 8th edition) Where final outcome depends primarily upon combination
of strategies selected by adversaries.
Emphasis on decision‐making processes of adversaries
We will focus on simplest case:
two‐person, zero‐sum games
Optimization & Decision 2009 494
The odds and evens game Two‐person, zero‐sum game
Player 1 takes evens, player 2 takes odds
Characterized by:
Each player simultaneously shows 1 or 2 fingers Strategies of player 1;
Player 1 wins if total of fingers is even and loses if it is Strategies of player 2;
odd; vice‐versa for Player 2 Payoff table.
E h l
Each player has 2 strategies: which?
h t t i hi h? Strategy: predetermined rule that specifies
Payoff table: completely how one intends to respond to each
Player 2 (odd)
possible circumstance at each stage of game
Strategy Payoff table: shows gain (positive or negative) for
1 2
1 1 ‐1 one player that would result from each combination
Player 1
(even) of strategies for the 2 players.
2 ‐1 1
Optimization & Decision 2009 495 Optimization & Decision 2009 496
Game theory Prototype example
1
Formulation Variation 1 of example
Total Net Votes Won
Identify the 2 players, the strategies of each player Given the by Politician 1
payoff table, (in Units of 1,000 Votes)
and the payoff table
Politician 2
Each player has 3 strategies: which strategy
should each Strategy 1 2 3
1. Spend 1 day in each city
p y y 1 1 2 4
player select?
l l t?
2. Spend 2 days in Bigtown Politician
2 1 0 5
1
3. Spend 2 days in Megalopolis
3 0 1 ‐1
Appropriate entries for payoff table for politician 1
are total net votes won from the opponent resulting Apply concept of dominated strategies to rule out
from 2 days of campaigning. succession of inferior strategies until only 1 choice
remains.
Optimization & Decision 2009 499 Optimization & Decision 2009 500
Dominated strategy Variation 1 of example (cont.)
A strategy is dominated by a second strategy if the Strategy 3 for player 2 is now dominated by
second strategy is always at least as good (and strategies 1 and 2 of player 1.
sometimes better) regardless of what the opponent Reduced table: 1 2
does. A dominated strategy can be eliminated 1 1 2
immediately from further consideration.
y 2 1 0
Payoff table includes no dominated strategies for Strategy 2 of player 1 dominated by strategy 1.
player 2.
Reduced table: 1 2
For player 1, strategy 3 is dominated by strategy 1. 1 1 2
Resulting reduced table: 1 2 3
1 1 2 4 Strategy 2 for player 2 dominated by strategy 1.
2 1 0 5 Both players should select their strategy 1.
Optimization & Decision 2009 501 Optimization & Decision 2009 502
Value of the game Variation 2 of example
Given the payoff table, which strategy should each
Payoff to player 1 when both players play optimally is player select?
value of the game. Saddle point
Total Net Votes Won
Game with value of zero is a fair game. by Politician 1 (equilibrium
(in Units of 1,000 Votes) solution)
Concept of dominated strategy is useful for:
Politician 2
Reducing size of payoff table to be considered;
Strategy 1 2 3 Minimum
Identifying optimal solution of the game (special cases). 1 ‐3 ‐2 6 ‐3
Politician
2 2 0 2 0 Maxmin value
1
3 5 ‐2 ‐4 ‐4
Both politicians
Maximum 5 0 6
break even:
Minimax value
fair game!
Optimization & Decision 2009 503 Optimization & Decision 2009 504
2
Minimax criterion Variation 3 of example
Each player should play in such a way as to minimize Given the payoff table, which strategy should each
his maximum losses whenever the resulting choice of player select?
strategy cannot be exploited by the opponent to then Total Net Votes Won Cycle!
improve his position. by Politician 1
Unstable solution
(in Units of 1,000 Votes)
Select a strategy that would be best even if the Politician 2
selection were being announced to the opponent Strategy 1 2 3 Minimum
before the opponent chooses a strategy. 1 0 ‐2 2 ‐2 Maxmin value
Politician
Player 1 should select the strategy whose minimum 1
2 5 4 ‐3 ‐3
payoff is largest, whereas player 2 should choose the 3 2 3 ‐4 ‐4
one whose maximum payoff to player 1 is the smallest. Maximum 5 4 2
Minimax value
Optimization & Decision 2009 505 Optimization & Decision 2009 506
It is necessary for each player to use one of her pure Suppose politicians 1 and 2 select the mixed strategies
strategies. (x1, x2, x3) = (½,½,0) and (y1, y2, y3) = (0,½,½).
Pure strategy would be chosen by using some Each player could then flip a coin to determine which of
random device to obtain a random observation from his two acceptable pure strategies he will actually use.
p p g y
the probability distribution specified by the mixed Useful measure of performance is expected payoff:
strategy. m n
This observation would indicate which particular pure Expected payoff for player 1 = ∑∑ pij x i y j
strategy to use. i =1 j =1
pij is payoff if player 1 uses pure strategy i and player 2
uses pure strategy j.
Optimization & Decision 2009 509 Optimization & Decision 2009 510
3
Expected payoff (cont.) Minimax criterion for mixed strategies
4
Optimal solution for politician 1 Optimal solution for politician 2
ν =ν = max{min{−3 + 5 x 1 ,4 − 6 x 1 }}
0 ≤ x1 ≤1 Expected payoff resulting from optimal strategy for
x 1* = 7 all values of x1 satisfies:
11
y 1* (5 − 5 x 1) + y2* (4 − 6 x 1) + y 3* (3 − 5 x 1) ≤ν =ν = 2
x =4
*
11
2 11
ν =ν = 2 x1 = 7
When player 1 is playing optimally, and
11 11
20 y 1* + 2 y2* + 2 y 3* =ν = 2
11 11 11 11
Also y 1* + y2* + y 3* = 1
So y 1* = 0, y2* = 5 and y 3* = 6
Minimum expected 11 11
payoff
Optimization & Decision 2009 517 Optimization & Decision 2009 518
If there should happen to be more than two lines Any game with mixed strategies can be transformed
passing through the maximin point, so that more to a linear programming problem applying the
than two of the yj* values can be greater than zero, minimax theorem and using the definitions of
this condition would imply that there are many ties maximin value ν and minimax value ν .
for the optimal mixed strategy for player 2.
Set all but two of these yj* values equal to zero and Define ν = x m +1 = y n+1
solve for the remaining two in the manner just
described. For the remaining two, the associated
lines must have positive slope in one case and
negative slope in the other.
Optimization & Decision 2009 519 Optimization & Decision 2009 520
5
Duality Still a loose end…
Example Extensions
Consider again variation 3 Maximize x 3 Two‐person, constant‐sum game: sum of payoffs to
after dominated strategy 3 two players is fixed constant (positive or negative)
subject to
for player 1 is eliminated regardless of combination of strategies selected
5 x1 − x3 ≥ 0
Adding x3 0 yields N‐person game, e.g., competition among business
−2 x 1 + 4 x2 − x 3 ≥ 0 firms international diplomacy etc
firms, international diplomacy, etc.
x 1* = 7 , x2* = 4 , x 3* = 2
11 11 11 2 x1 − 3 x2 − x 3 ≥ 0 Nonzero‐sum game: e.g., advertising strategies of
Dual problem yields (y4 0) competing companies can affect not only how they
x1 + x2 = 1
will split the market but also the total size of the
y = 0, y = 5
* *
,y =6
*
, and market for their competing products. Size of mutual
1 2 11 3 11
x 1 ≥ 0, x2 ≥ 0 gain (or loss) for the players depends on combination
y =2
*
4 11 of strategies chosen.
Optimization & Decision 2009 525 Optimization & Decision 2009 526
Extensions (cont.) Conclusions
Nonzero‐sum games classified in terms of the degree
to which the players are permitted to cooperate General problem of how to make decisions in a
competitive environment is a very common and
Noncooperative game: there is no preplay
important one
communication between players
Fundamental contribution of game theory is a basic
g
Cooperative game: where preplay discussions and
conceptual framework for formulating and analyzing
binding agreements are permitted
such problems in simple situations
Research is continuing with some success to extend
Infinite games: players have infinite number of pure the theory to more complex situations
strategies available to them. Strategy to be selected
can be represented by a continuous decision variable
Optimization & Decision 2009 527 Optimization & Decision 2009 528