On cooperative games, inseparable by semivalues (Q1423677)

Semivalues constitute a family of allocation rules for cooperative games on a given player set \(N\) that include the Shapley and Banzhaf values. It is known that a semivalue may well assign the same payoff vector to different games. Even more, two different games may receive a common allocation under all semivalues. Due to additivity, this occurs if, and only if, the difference game belongs to the shared kernel, introduced as the intersection of the kernels of all semivalues and denoted by \(C_N\). Then, two games can be discriminated by means of some semivalue if, and only if, their difference is not a member of the shared kernel. Moreover, given a game \(v\), the set of games getting, through any semivalue, the same allocation as \(v\) form an additive coset \(v + C_N\) of the shared kernel in the linear space of games on \(N\). A system of linear conditions characterizing the games of the shared kernel, as well as the dimension of this linear subspace and an explicit basis, formed by the so-called shuffle games, are provided. Thus, for a given game \(v\) on \(N\), the amount \(a_{i,s}(v)\), for any player \(i\) and any cardinality \(s\) between 1 and \(n\), is defined as the sum of worths of all coalitions of size s containing player \(i\). These amounts are especially interesting since: (a) they are the only we need to compute any semivalue on \(v\); (b) game \(v\) belongs to \(C_N\) if, and only if, all these amounts vanish; and (c) games \(v\) and \(v'\) are inseparable by semivalues if, and only if, their corresponding amounts coincide for each \(i\) and each \(s\). It also follows that the dimension of \(C_N\) is \(2^n-n^2+n-2\). In particular, the shared kernel is null for \(n = 2\) or 3: in these cases, inseparable games do not exist. The caliber of any game is defined as the number of coalitions with nonnull worth in the game. The caliber of a game in the shared kernel is 0 or \(\geq 4\). Then, for \(n \geq 4\), a shuffle game \(v_{S,i,j,k,h}\) is defined in terms of a coalition \(S\), two members \(i\) and \(j\) of \(S\) and two members \(k\) and \(h\) of the complement set \(N \setminus S\). Precisely, \(v_{S,i,j,k,h} = 1_S + 1_{S\cup\{k,h\}\setminus\{i,j\}} - 1_{S\cup\{k\}\setminus\{i\}}- 1_{S\cup\{h\}\setminus\{j\}}\), where for any coalition \(T\), \(1_T\) denotes the unity game defined by \(1_T(T) = 1\) and \(1_T(R) = 0\) otherwise. All shuffle games are in the shared kernel and span it; however, for getting a basis only a subset of them is needed. In order to specify the members if this subset, the authors assume without loss of generality that the players are ordered, what allows them to speak of first, last and absent player of a coalition and consecutive coalitions. (More technical details are, of course, omitted in this report.)