This action might not be possible to undo. Are you sure you want to continue?

Welcome to Scribd! Start your free trial and access books, documents and more.

Find out moreOverview

Students will learn about • • • • • Introductory Examples Basic Definitions of Game Theory Basic Theorems 2xn Matrix Games Reducing of Game Size

**Introduction and Example
**

The theory of games analyzes competitive phenomena and seeks to provide a basis for rational decision-making. Its growing importance was highlighted in 1994 when the Nobel Prize in Economics was awarded to John Harsanyi, John Nash, and Reinhard Selten, for their pioneering work in the theory of non-cooperative games. The games in this section are matrix games whose various outcomes are listed in a payoff matrix. Two players in a game compete according to a fixed set of rules. Player R (for row) has a choice of m possible moves (or choices of action), and player C (for column) has n moves. By convention, the payoff matrix A = [aij ] lists the amounts that the row player R wins from player C, depending on the choices R and C make. Entry aij shows the amount R wins when R chooses action i and C chooses action j. A negative value for aij indicates a loss for R, the amount R has to pay to C. The games are often called two-person zero-sum games because the algebraic sum of the amounts gained by R and C is zero. EXAMPLE 1 Each player has a supply of pennies, nickels, and dimes. At a given signal, both players display (or “play”) one coin. If the displayed coins are not the same, then the player showing the higher-valued coin gets to keep both. If they are both pennies or both nickels, then player C keeps both; but if they are both dimes, then player R keeps them. Construct a payoff matrix, using p for display of a penny, n for a nickel, and d for a dime. Solution Each player has three choices, p, n, and d, so the payoff matrix is 3×3:

Consider a row for R and fill in what R receives (or pays), depending on the choice C makes. First, suppose R plays a penny. If C also plays a penny, R loses 1 cent, because

149

(See Fig. the worst that can happen to R is to win a penny.) That is. Player C notes that every column contains a positive entry and therefore C cannot be certain of winning anything. R has computed 150 . Otherwise. By looking at the payoff matrix in Example 1. Also. Finally. The (1. what has each player done? Player R has found the minimum of each row (the worst that could happen for that play) and has chosen the row for which this minimum is largest. because C displays the higher-valued coin. because either C matches the nickel or shows the higher-value dime. whichever is shown by C. when R plays a dime. the players discover that some plays are better than others. If C plays either a nickel or a dime. R gains either a penny or a nickel. suppose R plays a nickel. So player C chooses to play a penny. R wins the dime from C because of the special rule for that case. R also loses 1 cent. when both players display a dime. R loses the nickel. From a mathematical point of view.the coins match. If C plays a penny. Both players know that R is likely to choose a row that has positive entries. Player R notes that every entry in row 3 is positive and chooses to play a dime. because R’s dime is of higher value. while C is likely to choose a column that has negative entries (a payment from R to C). No matter what C may do. R wins the penny. This information goes in row 1: Next. 1. which will minimize the potential loss. 1) entry is –1.

a large positive payment to R is worse than a small positive payment. The situation is not quite so simple in the next example. In Example 1. then I’ll play a penny so that I can win a nickel. Player C.” Once R has done this. looking at the column maxima (the greatest payment to R). then aij is called a saddle point. R begins to reason. then I’ll play a dime so that I can win a penny. R will choose to play a nickel. “If R is going to play a nickel. ⎡ ⎤ min ⎢ max aij ⎥ i ⎣ j ⎦ max min aij = min max aij = 1 i j j i Basic Definitions DEFINITION If the payoff matrix of a matrix game contains an entry aij that is both the minimum of row i and the maximum of column j . C begins to reason. nickels.Observe that for C. however. and dimes to play. as the game begins.” However. Player C has found For this payoff matrix [aij ]. but this time the payoff matrix is given as follows: If player R reasons as in the first example and looks at the row minima. when C starts to play a dime repeatedly. . EXAMPLE 2 Again suppose that each player has a supply of pennies. Thus C has found the maximum of each column (the worst that can happen to C for that play) and has chosen the column for which this maximum is smallest. Thus. R and C both continue to play a nickel. As long as both players continue to seek their best advantage. player R will always display a dime (row 3) and player C will always display a penny (column 1). Some games may have more than one saddle point. and so on. . After a while. C switches to a nickel (to win a nickel) and then R starts playing a nickel . It seems that neither player can develop a winning strategy. 151 . thereby maximizing the minimum gain (in this case a loss of 1). “If C is going to play a dime. will also select a nickel to minimize the loss to R. the entry a31 is a saddle point for the payoff matrix.

If the five sectors of the circle are unequal in size. then the areas of the various sectors sum to 1. but with a certain probability attached to each possible choice. n m 152 . The region is cut into pie-shaped sectors. while This means that neither player can play the same coin repeatedly and be assured of optimizing the winnings. of selecting the various plays in the matrix game. In fact. The corresponding strategy for R is specified by a vector in R that lists the areas of the five sectors. or probabilities. the strategy is called a pure strategy. player R will select each of the five plays about 1/5 of the time. Here is a way to imagine how player R could develop a strategy for playing a matrix game. Indeed. if there are five sectors of equal area and if the arrow is spun many times. If the area of the circle is taken as 1 unit. when each move is made at random. Player R gives the arrow an initial spin and waits for it to come to rest. This strategy is specified by the vector in R whose entries all equal 1/5. DEFINITION A probability vector in R is the set of all x in R whose entries are nonnegative and sum to one. the payoff matrix for the game in Example 2 does not have a saddle point. But is it possible to formulate some combination of plays that over the long run will produce an optimal return? The answer is yes (as Theorem later will show). The strategy space for player R is the set of all probability vectors in R . For instance. The position of the arrowhead at rest determines one play for R in the matrix game. then in the long run some game plays will be chosen more frequently than the others. Suppose that R has a device consisting of a horizontal metal arrow whose center of gravity is supported on a vertical rod in the middle of a flat circular region. one for each of the rows in the payoff matrix. any predictable strategy can be countered by the opponent. and the strategy space for player C is the set of all probability vectors in R . when the game is played many times. If one entry in a strategy is 1 (and the other entries are zeros).Mathematically speaking. A point in a strategy space is called a strategy. and these areas give the relative frequencies. Such an x has the form m m 5 5 Let A be an m×n payoff matrix for a game.

This fact connects the theory of convex sets to the study of matrix games. Suppose R uses strategy x and C uses strategy y. it can be shown that the probability is x1y1 that R chooses the first row and C chooses the first column. depending on the row R chooses and the column C chooses. each strategy x is a linear combination.. and the strategy space for C is an (n − 1)-dimensional simplex in R . That is. m In general.that is. E(x.em . e1 . The strategy space for R is an (m − 1)-dimensional simplex in R ... Suppose now that R and C are playing the m×n matrix game A = [aij ].. a point in the convex hull of the set of standard basis vectors. say x and if C were to discover this strategy. Over the course of many games.The pure strategies in R m are the standard basis vectors for R . A similar computation holds for each possible pair of choices that R and C can make. The sum of the expected payoffs to R over all possible pairs of choices is called the expected payoff. y). Let X denote the strategy space for R and Y the strategy space for C. each strategy is a convex combination of the set of pure strategies . More precisely. + xm em . the number E(x. Roughly speaking.. There are mn possible outcomes of the game. If R were to choose a particular strategy. x1e1 + x2 e2 + . of these pure strategies with nonnegative weights that sum to one. the expected payoff to R for this outcome is a11x1y1 for one game. then C would certainly choose y to minimize The value of using strategy x is the number v( x) defined by 153 . y) is the average amount that C will pay to R per game. when R and C play a large number of games using the strategies x and y.. respectively. where aij is the entry in the ith row and the j th column of A. where n m Since R plays the first row with probability x1 and C plays the first column with probability y1 and since their choices are made independently. of the game to player R for strategies x and y. e2 .

y is E ( x. v(x) is the minimum of the inner product of x with each of the columns of A. A strategy x for R is called optimal if v( x ) = vR . the value of strategy y to C is the maximum of the inner product of y with each of the rows of A. as That is. y ) = x Ay is a linear functional on T T the probability space Y . That is.. A strategy optimal if y for C is called optimal if v( y ) = vC .Since x A is a 1×n matrix. This value vR can be thought of as the most that player R can be sure to receive from C. 154 . equality may be written. x is optimal for R if E ( x. A similar analysis for player C. (A linear functional on Y is a linear transformation from Y into R. usually denoted by aj . Since the minimum is attained when y = ej for some j .. Equivalently. The pure strategies are the extreme points of the strategy space for a player). The number T vC . y ) attains its minimum when y is one of the pure strategies. DEFINITION. Of course.. defined by is called the value of the game to C. the mapping y → E ( x. y) may exceed vR for some x and y if C plays poorly. shows that a particular strategy y will have a value v(y) given by because ei A = rowi(A). with x in place of x . using the pure strategies for x.en . This is the least that C will have to lose regardless of what R may do. is called the value of the game to rowplayer R. y ) ≤ vC for all x in X. e1 . it can be shown that E ( x. From this. The number vR . y ) ≥ vR for all y in Y . E(x. independent of what player C may do. for C. Thus. e2 . . The goal of R is to choose x to maximize v(x). defined by with the notation as described above. Recall that Aej is the j th column of the matrix A.

E ( x. y) for each y in Y . y) for each individual x. Solution Compute Next. ej ) for 1 ≤ j ≤ 3. y) ≤ v(y). y∈Y which proves the theorem. Thus E(x. vR ≤ vC . y ) ≤ min v( y ) . v(x) is the minimum of E(x. Thus. Compute E(x. y ) implies that v(x) ≤E(x. For any fixed y. y ) x∈ X Similarly. since v(y) is the maximum of E(x. Also. y) ≤ v(y) for all x from X and for all y from Y . y) and verify that this number lies between v(x) and v(y). for each x. y∈Y PROOF.Basic Theorems THEOREM In any matrix game. the left inequality above implies that v( x) ≤ max E ( x. y) over all x. So compute Then Similarly. For any x in X. 155 . EXAMPLE 3 Let where A comes from Example 2. These two inequalities show that v(x) ≤ E(x. v(y) ≥ E(x. the definition v ( x ) = min E ( x. as expected.

there are always optimal strategies. When ˆ ˆ ˆ ˆ ˆ ˆ ( x. y ) is called a solution to the game. DEFINITION The common value v = vR = vC is called the value of the game. there exists ˆ y in Y such that According to the Minimax Theorem. THEOREM (Minimax Theorem) In any matrix game.In Theorem above. Suppose 156 . A fundamental result in game theory is that vR = vC . the proof that vR ≤ vC was simple. by von Neumann and Oskar Morgenstern. y ) ≤ v( y ) = vC which shows ˆ ˆ that E ( x. vR = vC = v. The first proof by John von Neumann in 1928 was technically difficult. so there exists a point x in X such that Similarly. It appeared in the classic 1944 book Theory of Games and Economic Behavior. That is. but this is not easy to prove. THEOREM (Fundamental Theorem for Matrix Games) In any matrix game. y ) = v. y ) is a solution to the game. That is. an optimal row strategy and vR are fairly easy to compute. The next theorem is the main theoretical result of this section. vR = vC. Any pair ˆ ˆ of optimal strategies ( x. 2xn Matrix Games When a game matrix A has 2 rows and n columns. A proof can be based either on the Minimax Theorem or on the theory of linear programming (next section). every matrix game has a solution. Perhaps the bestknown proof depends strongly on certain properties of convex sets and hyperplanes. The proof based on the Minimax Theorem goes as follows: The function v(x) is ˆ continuous on the compact set X. vR = v( x) ≤ E ( x.

EXAMPLE 4 Consider the game whose payoff matrix is a. For instance.The objective of player R is to choose x in R to maximize v(x). a2 j ) . z) on the graph of v(x(t)). Notice that the line goes through the points (0. sketch the four lines and darken the line segments that correspond to the graph of z = v(x(t)). Identify the highest point M = (t. 2). the line z = 6 · (1 − t) + 2 · t for column 4 goes through the points (0. Thus v(x(t)) is the minimum value of n linear functions of t . The heavy polygonal path in Fig. a1 j ) and (1. 6) and (1. 2. the probability space X for R may be parameterized by a variable t . The z-coordinate of M is the ˆ value of the game for R. 2. b. and the maximum value of v(x(t)) is easy to find. the graph of z = v(x(t)) as a function of t becomes evident. with a typical x in X having the form 2 for 0 ≤ t ≤ 1. 2 represents v(x) as a function of t . Then v(x(t)) is the minimum of the inner product of x(t) with each of the columns of A. and the t-coordinate determines an optimal strategy x(t ) for R. Since x has only two entries. 157 . On a t -z coordinate system. because the z-coordinate of a point on this path is the minimum of the corresponding z-coordinates of points on the four lines in Fig. That is. When these functions are graphed on one coordinate system for 0 ≤ t ≤ 1. Solution a) The four lines are See Fig. The process is illustrated best by an example.

. + yn en . as in Fig. The highest point. ˆ satisfies the equation In addition. ˆ for each i such that xi PROOF.. M.. the coordinates of M are (2/5. . The next theorem supplies the key information about this. So the optimal strategy for R is For any 2×n matrix game. With this value available. n. The value of the game for R is 11/5. an analysis of the graphical solution for R. So there exist nonnegative numbers εj such that Then 158 . 2. 11/5). y ) = v. will reveal how to produce an optimal strategy for C. ˆ THEOREM Let x and v. Write ˆ ˆ ˆ ˆ ˆ ˆ ˆ ˆ y = y1e1 + y2 e2 + . e j ) for j = 1.b. Example 4 illustrates the method for finding an optimal solution for player R. and suppose that Then ˆ y be optimal strategies for an m×n matrix game whose value is ˆ ˆ y is a convex combination of the pure strategies ej in R n for which E ( x. y ) = v( x) ≤ E ( x. and the value of the game is the same for C as for R. Solve the equations for columns 1 and 3 simultaneously: Hence. on the graph of v(x) is the intersection of the lines corresponding to the first and third columns of A.. y ≠ 0. and this value is attained at 2/5. Theorem guarantees that there also exists an optimal strategy for player C. . and note that v = E ( x. .

ei ) = v. ˆ ˆ E ( x. which means that while Thus. 4. . ˆ ˆ Thus y is a linear combination of the ej for which εj = 0. This equality is possible only if y j = 0 whenever εj > 0. By Theorem above. . For such j . y ) ≤ v( y ) = numbers δi such that Then ˆ ˆ E ( x. observe that E (ei . ˆ ˆ Next. e4 ) are greater than 11/5 . e2 ) and E ( x. . ˆy) = v for each i such that x1 ≠ 0. EXAMPLE 5 The value of the game in Example 4 is 11/5 attained when Use this fact to find an optimal strategy for the column player C. . ˆ ˆ Then E (ei . ˆ and the t-coordinate identifies the optimal strategy x(2 / 5) = x . y ) for i = 1. 2 is the value of the game. ej ) for j = 1. E ( x. . the optimal ˆ column strategy y for C is a linear combination of the pure strategies e1 and e3 in R 3 . . 2 represent E(x(t ). y ) = v E(ei . So there exist nonnegative since the ˆ ˆ x1 sum to one. m. Only the lines for columns 1 and 3 pass through the point M. Solution The z-coordinate of the maximum point M in Fig.ˆ ˆ because the y j sum to one. ˆ y has the form 159 . This equality is possible only if δi = 0 when x1 ≠ 0. Recall that the z – coordinates of the lines in Fig.

. These observations are summarized in the following theorem. The optimal strategy for C is Reducing the Size of a Game The general m×n matrix game can be solved using linear programming techniques. Similarly. no matter what C may choose. in which case the dominating “larger” column is ignored. then let A1 be the (m − 1)×n matrix obtained by deleting row s from A. In either case. A similar analysis applies to the columns of A. any optimal strategy of the reduced matrix game A1 or A2 will determine an optimal strategy for A. In some cases. . Theorem 11 11 ˆ ˆ ˆ and E (e2 . THEOREM Let A be an m×n matrix game. c1 = 2 / 5 and c3 = 3 / 5 . however. . . y ) = . This means that for R the pure strategy of choosing row r is at least as good as the pure strategy of choosing row s. and for some choice by C. vector a is said to dominate vector b if ai ≥ bi for all i = 1. 160 . with entries ai and bi . if column t of matrix A dominates some other column. a matrix game can be reduced to a “smaller” game whose matrix has only two rows. and later we’ll discuss one method for doing this. and obtain 4c1 + (1 − c1 ) = 11/ 5 . DEFINITION Given a and b in Rn. let A2 be the m×(n − 1) matrix obtained by deleting column t from A. y ) = 5 5 For example. EXAMPLE 6 Use the process described in Theorem above to reduce the following matrix game to a smaller size. It follows that the recessive row s (the “smaller” one) can be ignored by R without hurting R’s expected payoff. Suppose that in the matrix game A. Substitute c3 = 1 − c1 . n and ai > bi for at least one i. respectively.where ˆ c1 + c3 = 1 . Since both coordinates of the optimal x are nonzero. Each condition. If a dominates b. then b is said to be recessive to a. If row s in the matrix A is recessive to some other row. If this happens. row r dominates row s. determines y . r is better than s. the graphical method of Examples 4 and 5 is available. Then find the value of the game and optimal strategies for both players in the original game. shows that E (e1 . by itself.

They intersect where t = 5/7. Delete row 2 and obtain This reduced 2×3 matrix can be reduced further by dropping the last column. So delete column 1 and obtain In this matrix. since it dominates column 2. player C will never want to use the first pure strategy. the original matrix game A has been reduced to and any optimal strategy for B will produce an optimal strategy for A. with zeros as entries corresponding to deleted rows or columns.Solution Since the first column dominates the third. Thus. Figure 3 shows the lines corresponding to the two columns of B. whose equations are z = 4t + 1 and z=−3t + 6. the value of the game is 2/7 and the optimal row strategy for matrix B is 161 . A quick check of matrix B shows that the game has no saddle point (because 3 is the max of the row minima and 5 is the min of the column maxima). row 2 is recessive to row 3. So the graphical solution method is needed.

162 . The first row is recessive to the third row. Deletion of the second and fourth columns leaves the matrix B: The game for B has no saddle point. compute The final step is to construct the solution for matrix A from the solution for matrix B (given above). but a graphical analysis will work. The two columns of B determine the two lines shown below. whose equations are 2·(1 − t) + 1·t and z=−1·(1 − t) + 2·t . and use the Theorem to write Solving gives Thus As a check. Look at the matrices to see where the extra zeros go. Example 7. respectively. respectively. the optimal column strategy must be a linear combi- ˆ nation of the two pure strategies. Find the optimal strategies and the value of the matrix game Solution. Set y = c1e1 + c2 e2 . The second and fourth columns dominate the first and third columns. The row and column strategies for A are. so the first row may be eliminated.Since the game has no saddle point.

As a check. satisfies two equations ˆ these equations determines y . the optimal column strategy y . compute This solves the game for B. The value of the game is 5/4 and the optimal row strategy for the matrix game B is ˆ By Theorem. Each of Thus.These lines intersect at the point (3/4. y ) = 5 / 4 . For example. Thus ˆ x for A needs a 0 in the first entry ˆ (for the deleted first row). 5/4). because x is a linear combination of both e1 and e2. the optimal column strategy y for A needs 0’s in entries 2 and 163 . y ) = 5 / 4 ˆ ˆ and E (e2 . The optimal row strategy 4 (for the two deleted columns). ˆ E (e1 .

y). paper”). Player R chooses one of the coins. two. and v(y) when x and y have the given values. scissors beats paper. v(x). C receives the sum of the two numbers. Find all saddle points for the matrix games in Exercises 5–8. 164 . a red 6. scissors. If the cards are of different colors. R pays N dollars to C. Player R has a supply of dimes and quarters. Player C has two cards: a red 4 and a black 9. write the payoff matrix for each game. at a given signal.Practice Problems In Exercises 1–4. In the case of a tie. or all five (paper). If the guess is correct. They each show one of their cards. Let M be the matrix game having payoff matrix Find E(x. the winner collects 5 yen. Players R and C each show one. In the case of a win. If N is odd. If the total number N of fingers shown is even. If the cards are the same color. 2. In the traditional Japanese children’s game janken (or “stone. 4. or three fingers. each of two players shows either no fingers (stone). and a black 7. 9. then C pays N dollars to R. and player C must guess which coin R has chosen. If the guess is incorrect. Player R has three cards: a red 3. there is no payoff. 1. C takes the coin. Stone beats scissors. and paper beats stone. R receives the larger of the two numbers. C gives R an amount equal to R’s chosen coin. 3. two fingers (scissors).

find the optimal row and column strategies and the value of each matrix game. but 7 army soldiers will be lost. Let M be the matrix game having payoff matrix Find E(x. If the convoy goes overland and encounters the guerrillas. the guerrillas can watch only one of the two roads. Each day a 165 . y). A certain army is engaged in guerrilla warfare. It has two ways of getting supplies to its troops: it can send a convoy up the river road or it can send a convoy overland through the jungle. v(x). and v(y) when x and y have the given values. 19.10. On a given day. If the convoy goes along the river and the guerrillas are there. the convoy will have to turn back and 4 army soldiers will be lost. In Exercises 11–18. half the supplies will get through.

what portion of the supplies gets through? 20. b. What is the optimal strategy for the army if it wants to maximize the amount of supplies it gets to its troops? What is the optimal strategy for the guerrillas if they want to prevent the most supplies from getting through? If these strategies are followed. what is the “value” of the game? What does this represent in terms of the troops? 166 . if the army encounters the guerrillas. In part (a). Thus. If it does not encounter the guerrillas. with R being the army. What is the optimal strategy for the army if it wants to minimize its casualties? What is the optimal strategy for the guerrillas if they want to inflict maximum losses on the army? If these strategies are followed. Find the optimal strategies for the army and the guerrillas with respect to the number of army casualties. Set up and solve the following as matrix games. a. there will be 9 casualties. whether they are attacked or not. what portion of the supplies gets through? b. there will be 2 casualties. a. the convoy gets through with no losses.supply convoy travels one of the roads. and if the guerrillas are watching the other road. Suppose in Exercise 19 that whenever the convoy goes overland two soldiers are lost to land mines.

- Permutation Matrices Whose Convex Combinations Are Orthostochastic
- Matrix
- Matrices - Ch. 1.8, 1.10
- Widrow HoffLearning LMS
- Matrix Algebra
- section5_3
- Answers to Problem Set # 8
- 7.1 Introduction to Matrices
- [J. Jones] Q.I.P
- 13 Deteminants & Matrices Part 2 of 6
- CE206_System of Linear Equations
- Linear Algebra Book
- Appendices
- Linear Algebra 01
- 11130146Chapter_2_-_Matrix
- Linear Algebra Midterm
- Systems of Linear Equations
- 6 1 Maxima and Minima
- 454
- chap3
- SciPy Tutorial
- Linear Algebra
- Maths M T.3 Game Theory
- Problem Set 15 Solutions
- Exercises in Linear Algebra_Erdman.pdf
- matrices.pdf
- MIT18_03S10_c08
- 02-lin-fcts
- maxima and minima of functions, how this concept is helpful in daily life.
- Cambridge sche

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.

scribd