N-Person Cooperative Games Strategic-Equilibrium by Gabriel J. Turbay, Ph.D.1 Keywords: Strategic-Equilibrium, Extended-Imputations, Utility-Transfer Analysis, Linearly-Balanced Collections, Theorems of the Alternative for Matrices ABSTRACT Based on von Neumann and Morgenstern´s detached extended imputations relation to the stable set solution of a cooperative game, necessary an sufficient conditions for the structural and strategic equilibriums that characterizes the symmetric (objective) solutions are given for all general-sum n-person cooperative games with transferable utility. The mathematical characterization of the equilibriums for these games is accomplished in terms of covering collections structures that are linearly-balanced, admissible utility transfers and fundamental theorems of the alternative for matrices: The Fredholm alternative for matrices form of the fundamental theorem of linear algebra and the Farkas lemma. The existence of a fundamental strategic equilibrium for every game is established and shown to constitute a von Neumann and Morgenstern un-dominated system of interrelated extended imputations. It is shown to be not necessarily a solution but a systemic attractor where all solutions may emerge from. Heuristic procedures to determine the fundamental equilibrium of a game and for generating vN-M non-discriminatory solutions are given. 1 Former Research Director, Facultad de Administración UNIVERSIDAD DEL ROSARIO, Bogotá Colombia -1- Introduction Except for vN-M analysis of the zero-sum and general sum 3-person cooperative games that identifies symmetric type solutions and a possible valuation for the players in the game, also found in the 4-person game; and the attempt to find all solutions to a general sum n-person cooperative game by means of the detached extended imputations, most of the existing solution concepts and criteria appear as forced solutions that result in disconnected versions of these symmetric solutions. When these solutions appear spread out in eventual partitions, a fragmentation emerges of what vN-M consider an integrated unit not to be taken in isolation as stated in “Theory of Games and Economic Behavior”2. There, the symmetric type solutions, not individually but as a set, not in actuality but in potentiality, are identified as the only strategic possibilities3 for the game. In the heuristic formulation of the stable sets or vN-M solutions, the definition captures in terms of the domination criteria, some essential aspects of the of the symmetric solutions stability; it does not focus however, on those properties that would allow to characterize and establish the necessary and sufficient conditions for the existence of such strategic equilibriums for 2 See vN-M 29.1.3 page 261 3 -2- general n-person cooperative games. This type of characterization is the objective hopefully to be accomplished in this paper. Basic concepts and definitions Let N = {1, 2,…, n} denote de set of n players and En denote the ndimensional Euclidian space. A game in characteristic function form is a pair = (N,v) , where v is a real-valued set function defined on the subsets of N that satisfy the following properties: (2.1) v() = 0 (2.2) v(S T) ≥ v(S)+ v(T) whenever S T =, S and T subsets of N (2.3) A game = (N,v) is said to be essential if v( N ) v({i}) iN Though essentiality is a desired property in every game, when answering the question: If the grand coalition does not form, which coalitions are likely to form? It may be considered that the grand coalition cannot be formed or that the value V (N) = max [V(S)] over S N (to maintain the supper-additive property). Somehow, as it will be shown later, it is convenient to replace the essentiality condition above with the more general condition: (2.4) 𝑣(𝑆) > ∑𝑖∋𝑆 𝑣({𝑖}) for every S in some covering C of N -3- A covering collection of N or simply a cover of N is a set C = {C1 , . . . , Ck } of subsets of N such that ∀ 𝑗 ∈ 𝑁 ∃ 𝐶𝑖 C such that j Ci . |C | = k is the cardinality of C . Clearly, 𝑘 ⋃ 𝐶𝑖 = 𝑁 𝑖=1 A payoff vector for coalition S is any distribution vector x En where x = (x1, , . . .,xn)t , where xj ≥ 0 if j S, xj =0 if j S Similarly, a payoff vector x En, for a coverC of N, in the game v is a column vector defined by xt = (x1, x2, . . . ,xn) such that for all S C ∑ 𝑥𝑗 = 𝑣(𝑆) 𝑗∈𝑆 A payoff vector for the players in N satisfies: group rationality if 𝑥(𝑁) = ∑ 𝑥𝑗 = 𝑣(𝑁) 𝑗∈𝑁 coalitional rationality if for all S N 𝑥(𝑆) = ∑ 𝑥𝑗 ≥ 𝑣(𝑆) 𝑗∈𝑆 individual rationality if for all jN -4- 𝑥𝑗 ≥ 𝑣({𝑗}) An imputation for the game v is a payoff vector for N that satisfies both individual and group rationality conditions above. A coalition 𝑆 N is said to be efficient for a payoff vector x for S if 𝑥(𝑆) = ∑ 𝑥𝑗 = 𝑣(𝑆), 𝑗∈𝑆 An extended imputation4 is a payoff vector x for N that satisfies individual rationality. In a 0-normalized game a payoff vector is an extended imputation if and only if it is a non-negative vector. In the later this case the set of all extended imputations consist of the nonnegative orthant En+ An extended imputation is said to be detached (vN-M) if 𝑥(𝑆) = ∑ 𝑥𝑗 ≥ 𝑣(𝑆) 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑆 𝑁 𝑗∈𝑆 A cover C = {C1 , . . . , Ck } of N is said to be an efficient cover for x (an extended imputation x), or a covering support for x if for all 𝐶𝑖 C , 𝑥(𝐶𝑖 ) = ∑ 𝑥𝑗 = 𝑣(𝐶𝑖 ) 𝑗∈𝐶𝑖 4 The concept of extended imputation was introduced in vN-M p.364for analytical purposes. -5- An extended imputation x is said to be attainable if and only if there exist a cover C of N that is efficient for x. REMARK 1: Attainable-extended-imputations through a cover C = {C1 , . . . , Ck } are payoff vectors for the covering collection C of N in consideration. They may be thought of as the composition of k-different payoff vectors, x(1),. . . , x(k) , for coalitions C1 , . . . , Ck respectively. So that Ci is an efficient set for for x(i) , i =1,. . ., k. The vectors x(i) ,i=1,. . .,k are related to each other by the fact that if Cr and Cs are two different coalitions and their intersection is not empty, Cr Cs then for any j Cr Cs , the payoff to player j in x (r) equals the one in x (s). That is, xj(r) = xj(s) , whenever j Cr Cs . Extended-imputations with cover-support are not necessarily detached. Nor, detached 5 extended–imputations must have necessarily a cover support. Extended-imputations are not limited to the restriction imposed by group rationality. That is, one may have. 𝑥(𝑁) = ∑ 𝑥𝑗 ≠ 𝑣(𝑁) 𝑗∈𝑁 5 Here a rational extended-imputation is equivalent to a detached extended imputation, in vN-M p.370. However not all detached extended imputations are attainable and here we are interested in detached extended imputations with covering support. -6- An extended-imputation that satisfies individual and coalitional rationality but x(N) < v(N) is known as a pre-imputation Proposition 1 The set of all detached xd-imputations for the cooperative game = (N,v) is a convex polyhedral set bounded bellow . Here is referred to as the over-all rationality-set ℛ̂ ( ) = {𝑥𝜖𝐸 𝑛 |𝑥(𝑆) = ∑ 𝑗∈𝑆 Proof: The over-all rationality-set 𝑥𝑗 ≥ 𝑣(𝑆), for all 𝑆 𝑁}. results from the intersection of half spaces. Hence it is a convex polyhedral set. This set is bounded bellow since cooperative game in characteristic function form determines a unique6 “vN-M value” ||2 such that for all x in ℛ̂ ( ), | 𝑥(𝑁) ≥ 𝑣(𝑁) + | |2 . Since we are interested in the bargaining dynamics of the game previous to the formation of the grand coalition, for now, we will restrict our attention to the set of extended imputations that satisfy individual and coalitional rationality. We will refer to this set, simply, as the rationality-set, defined by 6 See TGEB pag. -7- ℛ( ) = {𝑥𝜖𝐸 𝑛 |𝑥 (𝑆) = ∑ 𝑗∈𝑆 𝑥𝑗 ≥ 𝑣(𝑆), for all 𝑆 ⊂ 𝑁}. Elements of the rationality set are referred to as rational-extended imputations. The properties of proposition 1 for the overall rationality set are applicable to the less restricted rationality set. The former is a subset of the later. If an extended-imputation x is rational and has an efficient covering support C, then x is said to be rational attainable. extended-imputations that have an efficient cover support, C So, all detached in = (N,v) are rational-attainable. Proposition 2 The set of all attainable extended-imputations for the cooperative game = (N,v) is the union of connected convex sets . This set will be referred to, as The Bargaining - field for the game = (N,v) and will be denoted by B ( ). Proof: The rational set is clearly the intersection of half spaces. Hence B ()is a convex polyhedral set. However the bargaining field is the intersection of the hyperplanes with gradients defined by the characteristic -8- vectors of the coalitions in covering collections of N . These intersections clearly may belong or not to the rationality- set We may restrict the negotiations to the set of rational-attainable extended imputations, denoted by r-B (). This restriction will allow us to see better the attractor like characteristics of the fundamental-strategicequilibrium that every cooperative games exhibits. However it eliminates too many possibilities where bargaining processes may begin. By requiring extended imputations to be in the rationality set all attainable extended imputations dominated by detached –attainable extended imputations are not considered as bargaining possibilities to begin negociation processes. Proposition 3 r-B ( ) R ( ). The extreme points of r-B ( ), are those of R () . Example 1: For the 3-person cooperative game given below, the ̅̅̅̅̅̅ bargaining field consist on the three line segments ̅̅̅̅̅̅ 𝐴 𝑋° , ̅̅̅̅̅̅ 𝐵 𝑋° and 𝐶 𝑋°. The extreme points of the rationality set the for 0-normalized game here, is given R ( ) = {x En+ | x1 +x2 10 , x1 + x3 10 and x2 +x3 6 }. Clearly the extreme points of R ( ): A, B, C and X° are those of r-B ( ). -9- 3-Person general-sum game v({i})=0, i=1,2,3 , v({1,2}) = 10, v({1,3}) = 8, v({2,3}) = 6, v({1,2,3}) = 12 x2 E (0, 12, 0) B (8, 6, 0) A (0, 10, 8) x2 + x3 = 6 X o (6, 4, 2) F(12, 0, 0) x1 + x 2 = 10 x1 C(10, 0, 6) D (0, 0, 12) x3 Figure 1: The rational-bargaining field r-B ( ) = ̅̅̅̅̅̅ 𝐴 𝑋° ∪ ̅̅̅̅̅̅ 𝐵 𝑋° ∪ ̅̅̅̅̅̅ 𝐶 𝑋° for the general-sum 3-person cooperative game above. The concept of rational- attainable extended-imputation is essential to identify and characterize the structural and strategic equilibriums, and most important the fundamental strategic equilibrium for the general-sum nperson cooperative game. Remark 2 Any attainable payoff vector that satisfies individual and coalitional rationality is a vector in the rationality- set. The converse is not - 10 - necessarily. That is any vector in the rationality set is not necessarily attainable Utility transfer analysis7 Utility transfers are defined as vectors in En. These are expressed basically as the difference of payoff vectors. That is for x and y any extended imputations the utility transfer t = (1, 2, … , n ) is defined as = y – x or y = x + . Utility transfers may not be viable or acceptable by the players involved, so it is necessary to have a well define structure of payoff s and transfers to understand their nature and behavior. Proposition 4 Let x be a rational attainable xd – imputations with cover support C . Then a extended-imputation y = x + also attainable through C Proof: , with x y is if and only if (C) = 0 for all C in C . y = x+ implies y(C) = x(C)+ (C). clearly , given x(C) = v(C) for all C in C , Then y(C) = v(C) if and if (C)=0 for all C in C. 7 Several concepts and results in this paper were previously developed by the author in his doctoral dissertation. Turbay Gabriel J., RICE University (1976). - 11 - If Rn, ≠0 and (C)=0 for all C in C we say that is a utility transfer admissible by C Corollary : Given x and y, x y , two extended-imputations with cover support C , then the utility transfer = y - x ≠ 0, is admissible by C. Characteristic vectors and matrices To characterize the stability that may take place in the process of attaining sustainable payoff claims and to find the type of utility transfers, if any, admissible by a cover C = {C1,…,Ck }, the following simple procedure may be followed: The characteristic matrix of a cover C . = {C1,…,Ck } in =(N,v), is a 0-1 matrix Wk x n whose rows Wi * i = 1,2,…,k, are the characteristic vectors of the coalitions Ci C . If Wi*.= ( wi1 , w i2 , … ,w iin ), the elements of each row i are given by wij = 1, if j Ci, and wij =0 otherwise, i = 1,…,n . It follows that the n-vector J t = (1,1 ,. . . , 1) is the characteristic vector of the set of all players N ={1,2, . . . ,n}. It is also the gradient of the hyper-plane that contains the imputations n-1 simplex I for the game. In general, a characteristic row-vectors of a coalition S of N is the transposed of the gradient of the hyperplane x(S) = v(S) of attainable payoffs through S. - 12 - Example 1. In the 3-person zero sum game C ={{1, 2}, {2,3}} is a covering support for both extended imputations : x = (1, 0, 1) y = ( 2/3, 1/3, 2/3). The utility transfer such that y = x+ and is given by = (- 1/3, 1/3, - 1/3). Clearly the utility transfer is admissible by C, , given that (C1 ) = 0 and (C2)= 0, for C1 = {1, 2}and C2= {2, 3}. Once the characteristic matrix corresponding to a cover C is constructed , the set of admissible transfers with respect to C , if any, may be obtained by solving for = ( 1, 2,… , n )t, the homogeneous system W = 0 . Player´s claims and bargaining alternatives If an extended imputation x is supported by a cover C the payoffs imputed by x are realizable through the coalitions in C . Given two different players h and k in a given coalition Cs C , we say that a coalition Ct C is a bargaining alternative of player h against player k if . h Ct and k Ct Bargaining alternatives represent a strategic basis to defend a particular - 13 - claim or stand prescribed in a given attainable extended imputation in the context of the cover C . In the example above, relative to x = (1, 0, 1), players 1 and 3 have no bargaining alternative defensible so the prescribed claims are not however relative to y = ( 2/3, 1/3, 2/3), player 2 in coalition {1, 2} has a bargaining alternative against player 1 and in coalition {2, 3} he has a bargaining alternative against player 3. As long as the players have bargaining alternatives against other coalition members that don´t, transfers of utility may be requested by the former players against the later ones. The admissible utility transfer structure in the example above is = (-, , - )t, The limit of this process relative to x is reached when the utility transfer is equal to = (- 1/2, 1/2, - 1/2) that gives x° = ( 1/2, 1/2, 1/2 ) at this point players 2 and 3 have bargaining alternatives that support the corresponding claims and the pair [x° , C ]8 with C = {{1, 2}, {2,3}, {1, 3}} becomes an equilibrium for the 3-person zero-sum game. The structure of the utility transfers, the bargaining field: the three line segments ̅̅̅̅̅̅ 𝐴 𝑥° , ̅̅̅̅̅̅ 𝐵 𝑥° and ̅̅̅̅̅̅ 𝐶 𝑥° , the bargaining alternatives {i, j} versus k i, j, k =1, 2, 3 and the equilibrium x° for the game are shown in the figure below: 8 All the stability arguments given by vN-M are applicable here see pag. - 14 - UTILITY TRANSFER STRUCTURE FOR THE 3-PERSON ZERO-SUM GAME x2 x3 = 0 (-, -, ) x1 = 0 B (1, 1, 0) A (0, 1, 1) x2 + x3 = 1 (, -, -) x1 x1 + x2 = 1 Xo (. 5, . 5, . 5) C(1, 0, 1) (-, , -) x2 = 0 x3 The following two propositions illustrate how the structure of supporting coalitions for extended imputations determine the structure of the admissible utility transfers, the bargaining alternatives for the players and hence the stability of the claims. Proposition 5 Let C = {C1,…,Ck } be a cover of N. If |C .| = k < n , then the cover C admits non-null utility transfers . - 15 - Proof: Clearly, any system of homogeneous equations W = 0 has a non trivial solution 0, whenever the number of equations k is less than the number of unknowns n. Corollary: The dimension of the admissible transfer space (C )is the same as the nullity of W. That is Dim[ (C )] = n-k The rank of the characteristic matrix W of a cover C . of N depends on the number of linearly independent rows of W. If the characteristic vectors of a collection of subsets of N are linearly independent we say that the corresponding coalitions are linearly independent subsets of N Proposition 6 Let C . = {C1,…,Cn } a covering collection of linearly independent subsets of N . Then there exist no utility transfer 0 admissible b: y the cover C . Proof: Since W has full rank, that is, rank (W) = n , the corresponding homogeneous system W = 0 has no non-trivial solution. Hence no admissible transfers are allowed by C . - 16 - In the example above, the collection of the three 2-person coalitions has a corresponding characteristic matrix with full rank. This implies the inexistence of admissible utility transfers relative to the equilibrium[x° , C ] . Such condition is reflected in the fact that any request made by one player to another one for a transfer in excess of the amounts prescribed by x° will not find a bargaining alternative. The requesting player will fail to get the additional amount from a third player because the third player would be able to protect his actual stand using the corresponding bargaining alternative, in which the requested player is included and may keep his actual stand but the requesting one is excluded Necessary and sufficient conditions for the admissibility of transfers are given by the fundamental theorem of linear algebra in the Fredholm alternative for matrices form. In our particular case it gives us the following: Proposition 7. (Utility transfer admissibility theorem) Given a cover C of N with characteristic matrix W; Either the system Wt = J has a solution Or the system W = 0 , Jt 0 has a solution - 17 - But not both can have a solution. Proof: The above theorem is simply an application of the well known fundamental theorem of linear algebra. Corollary If Wt = J has a solution and there exist 0 such that W = 0, then Jt = 0. Proof: If Jt = t W then Jt = t W = 0, since W = 0, Whenever the system Wt = J has a solution the corresponding cover of N is said to be a linearly-balanced collection of subsets of N. Proposition 8 Any linearly-balanced collection C of subsets of N is a covering collection for N Proof: Since t W = Jt has a solution then the elements column W*j cannot be all zeros for otherwise t W*j = 0,. Since t W*j =1, it follows that every j in N belongs to at least one coalition Ci in C .. LetC . = {C1,…,Ck } be a linearly- balanced cover C For all j N , Let D (j) = {Ci C | j Ci} , and D - 18 - hl = { Ci C | h Cj, l Cj}. of N. The following proposition gives us an important property of lbalanced collections Proposition 9 A necessary and sufficient condition for C to be l- balanced is that for all h, l N, h l (∗) ∑ 𝛾𝑖 = 𝐶𝑖 ∈ 𝔇ℎ𝑙 ∑ 𝛾𝑖 𝐶𝑖 ∈ 𝔇𝑙ℎ Proof: C is l- balanced if and only if for all j N (𝑖) ∑ 𝛾𝑖 𝐼ℎ (𝐶𝑖 ) = 𝐶𝑖 ∈ 𝔇(𝑗) ∑ 𝛾𝑖 = 1 𝐶𝑖 ∈ 𝔇(𝑗) . Hence for h l, h, l N we have that (𝑖𝑖) ∑ 𝛾𝑖 𝐼ℎ (𝐶𝑖 ) = 1 = 𝐶𝑖 ∈ 𝔇(ℎ) ∑ 𝛾𝑖 𝐼𝑙 (𝐶𝑖 ) 𝐶𝑖 ∈ 𝔇(𝑙) Let I = D (h) D (l) Both D (h)- [I D hl ]= and I D hl = Also D (l)- [I D lh ]and I D lh = Therefore - 19 - (𝑖𝑖𝑖) ∑ 𝛾𝑖 𝐼ℎ (𝐶𝑖 ) + 𝐶𝑖 ∈ ℐ ∑ 𝛾𝑖 𝐼𝑙 (𝐶𝑖 ) = ∑ 𝛾𝑖 𝐼ℎ (𝐶𝑖 ) + ∑ 𝐶𝑖 ∈ 𝔇(ℎ𝑙) 𝐶𝑖 ∈ ℐ 𝐶𝑖 ∈ 𝔇(𝑙ℎ) 𝛾𝑗 𝐼𝑙 (𝐶𝑖 ) =0 Then ( * ) follows when we subs tract ( iii ) from (ii ). Conversely if we can find numbers (not all zero) 𝛾1^ , … . , 𝛾𝑘^ Such that ( * ) hold true for all j in N then (i) must also hold true therefore we must have ∑ 𝛾𝑖^ 𝐼ℎ (𝐶𝑗 ) = 𝐶𝑖 ∈ 𝔇(ℎ) Defining ∑ 𝛾𝑖^ 𝐼𝑙 (𝐶𝑖 ) = 𝑏 ( 𝑏 𝑖𝑠 𝑎 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡) 𝐶𝑖 ∈ 𝔇(𝑙) 𝛾𝑗 = 1 𝑏 𝛾𝑗^ , (𝑗 = 1, … . . , 𝑘) we conclude that C must be l- balanced with balancing vector = (1, ……, k)T. The following definitions discriminate deeper than those of Bondareva ( ) and Shapley( ): A covering collection C = {C1, . . . , Ck} of subsets on N with characteristic matrix W, is said to be l-balanced if | W t = J, unrestricted (linearly-balanced) n-balanced if | W t = J, and i | i <0, i=1,. . . ,k (negatively balanced) nn-balanced if 0 | W t = J, (non-negatively-balanced) - 20 - Non-negatively- balanced9 coverings may be partitioned into the following two kinds, of covering collections: p-balanced if > 0 | W t = J, (positively-balanced) w-balanced if 0 | W t = J, and i = 0, for at least one element of . Rm is referred to as the balancing vector. The elements i of , i=1,…,m, are the balancing coefficients of the sets in C. . A utility transfer 0 is said to be a zero-sum utility transfer if and only if (i) W = 0 (ii) Jt = 0 9 (admissibility) ( pareto optimality) Here p-bal,anced corresponds to balanced and w-balanced to weakly-balanced in Shapley( ) , - 21 - Structural equilibrium Let C = {C1, . . . , Ck} be covering collection of subsets on N and X(C ) the corresponding set of rational-attainable extended imputations. Then The pair [ X(C ), C ] constitute a structural equilibrium for the game = (N,v) if and only if the only utility transfers admissible by C , if any, are zero-sum. Now we state the fundamental structural equilibrium theorem: Theorem I The pair [ X(C ), C ] as above constitute a structural equilibrium for the game = (N,v) if and only if the collection C is a linearly balanced collection of subsets of N. Proof: A direct consequence of the corollary to proposition 7 is that if the collection C is l-balanced the only possible transfers if any must be zerosum and conversely The definition of structural equilibrium in terms of utility transfers admissibility above implies that if the only admissible transfers by a cover C are zero-sum then it must be l-balanced. The following theorem gives us a special characteristic of balanced collections in terms of bargaining alternatives for the players: - 22 - Theorem II In any structural equilibrium [ X(C ), C ], for any extended imputation in X(C ) (of attainable claims), every player has a bargaining alternative versus any other non-symmetric player Proof: From proposition 9 we know ∑ 𝛾𝑖 𝐼ℎ (𝐶𝑖 ) = 𝐶𝑖 ∈ 𝔇ℎ𝑙 ∑ 𝛾𝑖 𝐼𝑙 (𝐶𝑖 ) 𝐶𝑖 ∈ 𝔇𝑙ℎ Hence either the players are symmetric if the sums above equal zero or each player has at least one bargaining alternative versus the other player. Strategic dependence Let h and l be different players. Relativc to a cover C of N we say that l is strategic-dependent on h if and only if D (h) D (l) .That is, if any coalition in C that contains h also contains l but the converse is not true. If D (h) =D (l) the players are symmetric and hence strategic- quivalent A Strategic-dependent players have no bargaining alternatives against the player hey depend from. Clearly, Proposition 10 If C is l-balanced, no player can be strategic-dependent on another player. - 23 - The strong structural stability exhibited10 by collections with n linearly independent vectors motivates an analysis of the rationality set: ℛ( ) = {𝑥𝜖𝐸 𝑛 |𝑥(𝑆) = ∑ 𝑗∈𝑆 𝑥𝑗 ≥ 𝑣(𝑆), for all 𝑆 ⊂ 𝑁} We know ℛ( ) is a convex polyhedral set. In matrix notation, ℛ( ) = { 𝑥| 𝑉𝑥 ≥ v }, where V ( 2n- 2 x n) is the characteristic matrix of the proper subsets of N and v the corresponding vector of characteristic function ̂ the characteristic values. Let 𝑥̂ an arbitrary extreme point of ℛ( ) and 𝑊 matrix of the coalitions in active at 𝑥̂ . Proposition 10 Any 𝑥̂ extreme in ℛ( ) with its corresponding covering support C conform a structural-equilibrium for the game = (N,v) Proposition 11. The extreme points of ℛ( ) are the basic solutions of the system 𝑉𝑥 ≥ v and hence, there are n linearly independent subsets of n active at any 𝑥̂ in ℛ( ) 10 All vN-M stability arguments applicable to the symmetric solution to the 3-person zero-sum game are applicable here see pp. - 24 - Example 2 : For the 3-person general sum game with characteristic function given by vi = 0 if i=1, 2, 3 , v({1, 2}) = 5, v({1, 3}) = 6 , v({2, 3}) = 17, v({1, 2, 3}) = 19, the extreme point of ℛ( ) are given by the points A, B , C, D. Observe that A and D have a full rank n-balanced covering support while B and C have a full rank a w-balanced cover support. The bargaining field B ( ) here is given by the set of points ̅̅̅̅̅ conformed by the union of the segments ̅̅̅̅̅̅ 𝐴𝐵 , 𝐵 𝐶 and ̅̅̅̅̅̅ 𝐶𝐷. Strategic Stability x2 x3 = 0 D (0, 1, 1) x1 = 0 W o = 1 0 0 WB = 1 1 0, 0 1 1 1 0 0 1 WC = 0 1 1, = 1 1 0 1 0 = 1 0 1 (-, -, ) 1 0 1 0 1 1 (0, 11, 6) C 1 0 0 = 1 0 1 1 1 0 0 1 -1 WD = 1 0 1, = 1 0 1 1 1 (0,-, ) (0, 8, 9) o x1 (0, 5, 12) B (0, ,-) A (5, 0, 12) x1 + x2 = 5 x3 (-, , -) 1 0 1 1 0 1 0 1 0 WA = 0 1 1 , 1 1 0 - 25 - -1 = 1 1 x2 = 0 Figure 2. Strategic instability in negatively balanced Extreme points Proposition 12 The rationality set is bounded below so that for any x in ℛ( ) 𝑥(𝑁) ≥ 𝑣(𝑁) + | |2 Here, | |2 is the vN-M value11 determined by any game so that a detached extended imputation exists with an excess e = x(N) – v(N) ≥ | |2 . Proof: Clearly the removal of the constraint x(N) ≥ 𝑣(𝑁) wil not affect the lower bound of the overall-rationality–set of Proposition 1 above. To understand the relationship between n-balanced w-balance and pbalanced collections we establish the following propositions related to dual coalitions. Given a subset S of N, the dual of a coalition S is the set S*= N-S. If Ws is the characteristic vector of S then the characteristic vector of S* is given 0 𝑖𝑓 𝑗𝑆 by Ws ∗ = (wj∗ ) = { 1 𝑖𝑓 𝑗 𝑆 Similarly , if D is a collection of subsets of N the dual of D is defined by 11 See theorem on page - 26 - D * = {D* = N-D | D D } Proposition 13. Let C = { C1, …, Ck} be an l-balanced collection of subsets of N with corresponding balancing vector 𝛾 = (𝛾1 , … . , 𝛾𝑘 )𝑇 , and let D be a proper sub-collection of C such that ∑ 𝛾𝑗 ≠ 1. 𝐶𝑗 ∈𝜏 ) D * is also l-balanced with Then the collection C ´ = (C - D balancing vector 𝛾𝑖 𝜌 𝛾 ´ = (𝛾1 ´ , … . , 𝛾𝑘 ´)𝑡 where 𝛾𝑖´ = { −𝛾𝑖 𝜌 Here 𝑖𝑓 𝐶𝑖 ∈ 𝒞´ 𝑖𝑓 𝒞𝑖 ∈ 𝔇∗ 𝜌 = 1 − ∑ 𝛾𝑗 𝐶𝑖 ∈𝔇 Proof: For any j N let D = {C ∈ τ j C} Since Cj 𝐶𝑗1 * = ( - D )* D then ∑ 𝛾𝑗´ 𝐼𝑗 (𝐶)= 𝐶𝑗 ∈𝒷 1 𝜌 ∑ 𝛾𝑗 𝐼𝑖 (𝐶𝑗 ) − 𝐶𝑗 ∈[𝒞− 𝜏] 1 𝜌 - 27 - ∑ 𝛾𝑗 𝐼𝑗 (𝐶)= 𝐶𝑗 [𝜏−𝒟]∗ 1 𝜌 ∑ 𝛾𝑗 𝐼𝑖 (𝐶𝑗 ) 𝐶𝑗∈𝒟 ∗ But i D , i D *, hence the right most term of ( ) vanishes and we obtain ∑ 𝛾𝑗^ 𝐼𝑗 (𝐶)= 𝐶𝑗 ∈𝒷 ^ 1 𝜌 𝛾𝑗 𝐼𝑖 (𝐶𝑗 ) − [ ∑ ∑ 𝛾𝑗 𝐼𝑖 (𝐶𝐽 )] 𝐶𝑗 [𝜏−𝒟]∗ 𝐶𝑗 ∈[𝒞−𝜏] Adding to the expression in brackets in ( ) ∑ 𝛾𝑗 𝐼𝑖 (𝐶𝐽 ) − ∑ 𝛾𝑗 𝐼𝑖 (𝐶𝐽 ) = 0 𝐶 𝑗∈𝒟 𝐶𝑗 ∈𝒟 And nothing that ∑ 𝛾𝑗 𝐼𝑖 (𝐶𝐽 ) = 𝐶𝑗 [𝜏−𝒟]∗ 1 𝜌 ∑ 𝛾𝑗 𝑎𝑛𝑑 ∑ 𝛾𝑗 𝐼𝑖 (𝐶𝐽 ) = ∑ 𝛾𝑗 𝐶 𝑗∈𝒟 𝐶𝑗 [𝜏−𝒟] 𝐶 𝑗∈𝒟 We obtain 1 ∑ 𝛾𝑗´ 𝐼𝑗 (𝐶)= 𝐶𝑗 ∈𝒷 ^ 1 𝜌 𝛾𝑗 𝐼𝑖 (𝐶𝑗 ) + ∑ 𝛾𝑗 𝐼𝑖 (𝐶𝐽 ) ) [( ∑ 𝐶𝑗 𝒟 𝐶𝑗 ∈[𝒞−𝜏] − ( ∑ 𝛾𝑗 + ∑ 𝛾𝑗 )] 𝐶𝑗 [𝜏−𝒟] 𝐶𝑗 ∈𝒟 - 28 - 1 𝜌 (1 − ∑ 𝛾𝑗 ) = = 1 𝜌 𝜌 𝐶𝑗 ∈𝜏 Which establishes that the collection C ´ is l-balanced with corresponding balancing vector ´. Corollary If C is p-balanced and D C , D , is such that 0 < ∑ 𝛾𝑖 < 1 𝑐𝑖 𝜖𝔇 Then C´´= (C - D ) D * is n-balanced with D * = {Ci C ´| ´i <0} and (C - D ) ={Ci C ´| i >0}. Proof : A particular case occurs when D = {Ch} then C *h = C – {Ch}{Ch*} is n-balanced, In this case i *h > 0 for Ci (C – {Ch}) and h*h < 0 Ch*}. Proposition 14 Any cover C´ ={ C1, …, Cn } of N consisting of n linearly independent subsets of N , is linearly balanced and has a full rank characteristic matrix Wn x n . For any ChC the collection C´´= C - { Ch} - 29 - has a characteristic matrix W´n-1x n .and admits utility transfers =[W-1h*] given by the hth column of W-1. For such transfers W 0, Wh*>0, Jt >0 Example 3 Consider the collection of subsets of N = {1, 2, 3, 4], where C = {C1, C2, C3, C4, C5} and C1 = {1, 2, 3}, C2 = {2, 3, 4}, C3 = {3, 4, 5}, C4 = {1, 4} and C5 = {1, 2, 5}. Here, C is p- balanced with characteristic matrix W and corresponding balancing g given by 1 0 W= 0 1 (1 1 1 0 0 1 2 −2 −1 3 4 3 −1 −2 2 [ 1 −1 1 1 1 0 0 0 1 1 1 0 1⁄4 0 1⁄4 0 1 and 𝛾 = 1⁄2 𝑊 −1 = 1/ 0 1⁄4 1) (1⁄2) 0 2 0 −2 −1 2 2 −1 −2 0 2 0 2 −1 2 ] Considering collections of the form C ´= [C - {Cn}] {C’n} (n= 1, 2, 3, 4, 5) we have that proposition 12 applies, and - 30 - 1 − 1⁄4 = 3⁄4 𝑖𝑓 𝜈 = 1, 2, 4 𝜌 = 1 − 𝛾𝜈 = { 1 − 1⁄2 = 1⁄2 𝑖𝑓 𝜈 = 3, 5 The coefficients of the balancing vector 𝛾̂ 𝛾̂𝑖 = 𝑖 𝑓𝑜𝑟 𝐶𝑖 {C - {Cn}}, 𝛾̂𝑖 = −𝑖 𝑓𝑜𝑟 𝐶𝑣 All collections C =C −{𝐶𝜈 } are structurally unstable and for any n = 1, 2, 3, 4 or 5 C ´ is n-balanced. Below we give the corresponding characteristic matrices, transfers and balancing vectors for n = 1, 2, 3, 4, 5. Characteristic matrix of Balancing vector C − {𝐶𝜈 } ̅𝜀 𝑊 Characteristic matrix of [C −{𝐶𝜈 } ] {𝐶 ∗𝜈 } Balancing Vector For n = 1 ∗∗∗∗∗ 00111 ̅ 𝑊 = 10010 11001 ( 1 1 0 0 1) − 0 ̅ 𝑊𝜀 = 0 0 (0) xT = (2, -, 3, -2,-) 00011 01110 ̂ 𝑊 = 00110 10010 ( 1 1 0 0 1) 𝐽t = For n = 2 - 31 - −1⁄3 1⁄3 𝛾̂ = 2⁄3 1⁄3 ( 2⁄3 ) ̂ = (1 1 1 1 1) 𝛾̂ T 𝑊 11100 ∗∗∗∗∗ ̅ 𝑊 = 00111 10010 (1 1 0 0 1) 0 − 0 0 (0) 11100 10001 ̂ 𝑊 00111 10010 ( 1 1 0 0 1) x t= (-2, 3,-, 2, -) 𝐽t = 1⁄3 −1⁄3 ̂𝛾 = 2⁄3 1⁄3 ( 2⁄3 ) ̂ = (1 1 1 1 1) 𝛾̂ T 𝑊 For n = 3 11100 01110 ̅ = ∗∗∗∗∗ 𝑊 10010 (1 1 0 0 1) 0 0 − 0 (0) 11100 01110 ̂ = 11000 𝑊 10010 ( 1 1 0 0 1) 𝐽 t = 2 xT = ( 0, -2, 2, 0, 2) 1⁄2 1⁄2 ̂𝛾 = −1 1⁄2 ( 1 ) ̂ = (1 1 1 1 1) 𝛾̂ T 𝑊 For n = 4 11100 01110 ̅ 𝑊 = 00111 ∗∗∗∗∗ (1 1 0 0 1) 0 0 0 − (0) xT = ( 2, - , - , 2 - ,) 11100 01110 ̂ 𝑊 = 00111 ̂𝛾 = 01101 ( 1 1 0 0 1) 𝐽t = 1⁄3 1⁄3 2⁄3 −1⁄3 ( 2⁄3 ) ̂ = (1 1 1 1 1) 𝛾̂ T 𝑊 For n = 5 11100 01110 ̅ = 00111 𝑊 10010 ( ∗∗∗∗∗ ) 0 0 0 0 (−) xT = ( 0, 2, - 2, 0, 2) 11100 01110 ̂ = 00111 𝑊 10010 ( 0 0 1 1 0) 𝐽 t = 2 - 32 - 1 ⁄2 1 ⁄2 1 = ̂𝛾 1 ⁄2 ( −1 ) ̂ = (1 1 1 1 1) 𝛾̂ T 𝑊 The structure of the admissible utility transfers in each case v =1, 2, 3, 4, 5 is obtained from the corresponding column v of W-1 according with proposition 14. ̂ in all cases is n-balance and only differs We note that the Matrix 𝑊 from W in that one of the rows has been replaced by its complement (dual coalition) ̂ above when n = 5 Had we began our transfer analysis say with 𝑊 11100 01110 ̂ 𝑊 = 00111 , 10010 ( 0 0 1 1 0) 1⁄2 1⁄2 1 = ̂𝛾 1⁄2 ( −1 ) 1 −1 0 1 0 0 2 0 0 −2 −1 1 ̂ 𝑊 = 1 −1 0 −1 2 2 −1 1 0 1 0 0 2 0 −2) ( 0 The type of admissible utility transfers that result when the coalition corresponding to v = 5 (the collection with negative balancing coefficient) removed, is given by xT = ( 0, -2, 2, 0, -2) . - 33 - is We may observe that the structure of the utility transfers is identical to the one of case v= 5 above, but in the opposite direction. By examining the n-balanced collections in example 2 above, we note that the coalitions with negative coefficients hinder the strategic possibilities of both players 2 and 3 since from extreme points A and D each player can leave the restriction that is binding him to a zero payoff and use the coalition structure where player 1 have no bargaining alternative. Remark 3: Note that the admissible utility transfers that may emerge when a coalition with negative balancing coefficient is removed are such that some or all the players in the removed coalition will benefit from the transfer since ̂5∗ > 0. However the transfers brings an overall value loss since 𝐽𝑡 < 0 . 𝑊 That is the transfer moves from an extended imputation of attainable claims x to another one y = x + with corresponding value levels y(N) < x(N). Here, ̂𝑖∗ = 0 𝑓𝑜𝑟 𝑖 ≠ 5, 𝑎𝑛𝑑 𝑊 ̂5∗ > 0. 𝑇ℎ𝑢𝑠, 𝑊 ̂ ≥ 0, 𝑊 𝐽𝑡 = −1 < 0 But this conditions according to the Farkas lemma occurs only when the collection is not nn-balanced. - 34 - Proposition 15 (Farkas Lemma) Given a cover C of N with characteristic matrix W, Either the system Wt = J 0 has a solution Or the system W 0, Jt < 0 has a solution But not both can have a solution. Corollary: If there exist a utility transfer admissible by some p-balanced sub-collection of a nn-balanced coverC such that W 0 then the utility transfer must be necessarily pareto-optimal. In brief, if a p-balanced collection F with |F |= k linearly independent subsets, k < n, by proposition 5, it must admit utility transfers. However, the corollary to proposition7 tells us that such admissible transfers must be pareto-optimal. So if 𝑥̂ is an extreme point of the rationality set, and C for ̂𝑥 and F C is the corresponding l-balanced cover , then C coefficients non-positive. - 35 - must have (n-k) of its balancing Strategic-equilibrium We have shown with examples that n-balanced covers are not strategically stable because removing the coalitions will not hinder the bargaining alternatives of players and on the contrary will give opportunity for admissible utility transfers that will benefit at least some of the players in the removed coalition so the players may not be willing to stick to such coalitions that act as restraints. This shows that an structural equilibrium may not be strategically stable. Such condition is clearly corrected when there exist no admissible utility transfers with W 0 and Jt < 0. That is, a pair [ X(C ),C ] is an strategic-equilibrium for the the game =(N,v) if the cover support C for x in X(C ) is a nn-balanced covering collection of N. Clearly we have a strong-strategic–equilibrium if our cover has pbalanced collection of n-linearly-independent coalitions. In this case X(C ) = {x°} This strong-equilibrium pair will be denoted simply - 36 - as [x°,C ]. If the corresponding nn-balanced collection happens to be w-balanced, the equilibrium is said to be a weak-strategic equilibrium. Thus, in any cooperative game we may have as many strong-strategic equilibriums as there are p-balanced collections of subsets of N consisting of n-linearly independent subsets of N. Clearly, Not all of these equilibriums may satisfy individual and coalitional rationality. If an strategic-equilibrium is in the rationality set we say that such equilibrium is a fundamental strategic-equilibrium for the game = ( N, v ). Proposition 6 Any extreme point 𝑥̂ in ℛ( ) has negatively balanced cover support unless 𝑥̂ = x° where Jt x°= min Jt x subject to 𝑉𝑥 ≥ v . Proof: the dual to the above linear program is given by max t v subject to tV = Jt , 0 The primal and dual problems are convex polyhedral sets bounded below and above respectively. Hence both primal and dual problem have optimal solutions12 x°, ° respectively. Further, Jt x° = °tv with ° 0 . Any other 12 See Owen G.(1982) p - 37 - extreme solution 𝑥̂ to the minimization problem must have 𝑥̂ (N) = x°(N) and the corresponding balancing vector ̂ must be non negative . As corollary to proposition 16 , we have the following theorems. THEOREM III The pair [ X(C ),C ] constitute a strategic-equilibrium for the game = (N, v) if and only if the corresponding covering collection C of N is nn--balanced . THEOREM IV (Existence) Every cooperative game =(N,v) has a fundamental strategic equilibrium. All linearly defined systems are relative invariants under strategic equivalence. THEOREM V (Invariance) Any equilibrium (structural or strategic) for a game = (N, v) is a relative invariant under strategic equivalence. The above theorems are consequences of the LP characterization our equilibriums . Specific proofs are given in 13 13 Turbay G.J.(1976) “On value theories for N-person cooperative games”, Doctoral Dissertation RICE UNIVERSITY. - 38 - Clearly any extreme point of the rationality set (basic solution) has a l-balanced cover support. If the l-balanced cover support for 𝑥̂ happens to be nn-balanced then 𝑥̂ must be an optimal solution to the primal and the corresponding non-negative balanced vector ̂ 0 must be a solution to the dual problem. Given an nn-balanced linearly independent cover C of N with |C | = n , and let C |C + + = { Ci | i > 0 } and C 0 = { Ci | i = 0 } so that | + |C 0 |=n. If a feasible extreme extended imputation ̂𝑥 for the primal problem has a nn-balanced cover C with | C | = n and the number of positive coefficients p = | C + | = n, then x° = 𝑥̂ is the unique solution to the primal problem. If 0 < p <n then |C 0 | = n -p > 0 , the dimension of the transfer space is (n-p) and the number of optimal extreme points is (n-p) and the equilibrium is a weak-strategic equilibrium. Remar k 4: The strategic-equilibrium for any cooperative game may be obtained by using the dual simplex method for LP problems. However, as with the covering problem in integer programming the nature of the constraint gives high degrees of degeneracy. - 39 - A heuristic procedure for finding the fundamental-strategic equilibrium of a game is to place in descending order the coalitions according to the perplayer- characteristic function value (S) = 𝑣(𝑆) |𝑆| and form groups of n linearly independent nn-balanced sets with the higher possible values for (S) and obtain the corresponding unique supported extended imputation. If we can produce one that satisfy individual and coalitional rationality, such extended imputation and corresponding cover support constitutea fundamental strategic equilibrium for the game. “Systems thinking” view of the strategic-equilibrium The fundamental strategic equilibrium identified for cooperative games is not necessarily a solution to the game. It may be thought of as a fundamental attractor from which all solutions to the game may emerge. The following views can be traced to vN-M heuristic “systems thinking” views in the TGEB: In relation to the symmetric solution of the 3-person zero sum game: - 40 - (1) Solutions are not simply outcomes, but a system of interrelated outcomes; none of them, by itself, is to be considered a solution. (2) The outcomes in the solution are not events but “possible events”, so solutions are conditional interrelated sets of outcomes. In relation to the discriminatory solutions: (3) These solutions were never thought to exist in advance, they simply emerge as sets that satisfy the stability conditions : “… these situations arose even in the extremely simple framework of the zero-sum….” Regarding composition and decomposition of extended imputations: (4) brings out the possibility of “viewing as one” two separate occurrences” The above four basic points are reflected in our approach mainly as follows: - 41 - Rational-attainable extended-imputations with their corresponding covering support represent conditional systems. Through one such extended-imputation we are viewing as “one” different possible occurrences The possible occurrences are related among them by the fact that they represent bargaining –alternatives: That is, mutually exclusive possibilities for the players that may be realizable. In each interrelated alternative the players can obtain the bargaining claim prescribed by the extended imputation in consideration. Example 4: Consider a 4–person game with characteristic function given by 200 if |S| = 4 v(S) = 120 if |S| =3 and player 1 is not in S 120 if |S| =2 and player 1 is in S 0 otherwise The extended-imputation x° = (80, 40, 40, 40) t is rational attainable extended imputation with p-balanced cover support given by - 42 - C = { {1, 2}, {1, 3}, {1, 4}, {2, 3, 4}}. And corresponding balancing vector ° = (1/3, 1/3, 1/3, 2/3) t . hence the pair [x°, C ] is a fundamental-strong equilibrium the game . It represents the following conditional system of interrelated possible occurrences: (80, 40, 0, 0) if {1,2} occurs (80, 0, 40, 0) if {1,3} occurs (80, 0, 0, 40) if {1,4} occurs ( 0, 40, 40, 40) if {2, 3, 4} occurs The fundamental strategic-equilibrium is a pair [x°, C ] that all0w us to “view as one”, one extended imputation supported by one cover structure that describes four possible occurrences. The extended imputation x° is actually the composition of four payoff vectors (extended imputations) for the four corresponding possible coalitional eventualities. Strategic-equilibrium and vN-M non discriminatory solutions Continuing with the example above, if the players in the equilibrium cover form syndicates, they could realize binding agreements so as to bargain - 43 - with the excluded players for is incremental value contribution if the grand coalition is to form. Thus, If players 1 and 2 form a syndicate [1, 2] they may possibly agree: (1) To form a syndicate so as to secure the characteristic function value to the coalition v{1, 2} = 120 (2) To split the 120 as prescribed by x°= (80, 40, 40, 40) t and use the corresponding “syndicate dividends” as disagreement payoffs for coalition {1,2} d = (80, 40, 0, 0) (3) To split with coalition {3, 4} their incremental contribution ( if the grand coalition is to form ) e34= v(N) - v({1, 2}). So that a syndicate external division rate must be established with the excluded coalition {3,4} in the negotiation of e34. (4) To split the proceeds of the syndicate negotiations according to a prescribed syndicate internal division rate. The possible outcomes that may occur if the syndicate [1, 2]forms and the corresponding binding agreements are accepted by the players, are given by the following conditional system described by a generic row vector of imputations : - 44 - If [3, 4] form a counter-syndicate with internal splitting rate x1 80 + e34 With x2 x3 40 (1-) e34 x4 (1-) e34 (1-)(1-)e34 0< , , <1 If players 3 and 4 negotiate separately we obtain: x1 80 + (1- (3 +4 )) e34 With x2 40+(1-) (1-(3 +4 ))e34 x3 x4 3 e34 4 e34 0< , , 3, 4 <1 We could continue the same procedure for all possible syndicates. The above constructive description of obtaining possible outcomes14, when applied to the general sum 3-person cooperative game, the possible outcomes description emerge as vN-M non-discriminatory “objective” solutions. This is the subject of forthcoming papers entitled “Rethinking the solutions to the general-sum 3-person cooperative game” and “ The Stronger Player Paradox” 14 - 45 - References Bondareva O. N Farkas Fredholm Hichmann and Hirsh Owen G Shapley L. Turbay G. J. Von Neuman J. and O. Motgenstern - 46 -