Representation of the core of convex measure games via Kantorovich potentials (Q2581794)

The paper generalizes a characterization of the core of convex measure games with an ultramodular characteristic function. It disposes of the need to assume that the underlying measurable structure is that of a standard Borel Space. It does so by taking an approach based on rearrangement techniques and the notion of Kantorovich potentials. \(\Omega\) denotes the set of players and a \(\sigma\)-algebra, \(\mathcal{A}\), of subsets of \(\Omega\), the set of possible coalitions. A \textit{convex} game is a set function \(\nu: \mathcal{A\rightarrow } \mathfrak{\mathbb{R}}\) such that \(\upsilon \left( \emptyset \right) =0,\) satisfying, for all \(\left( A,B\right) \in \mathcal{A\times A}\), \(\upsilon \left( A\cup B\right) +\upsilon \left( A\cap B\right) \geq \upsilon \left( A\right) +\upsilon \left( B\right)\). A set function \(m: \mathcal{A\rightarrow\mathbb{R}}\) is \textit{additive} if for every pair of disjoint measuarble sets \(\left( A,B\right) \in \mathcal{A\times A}\), \(m\left( A\cup B\right) =m\left(A\right) +m\left( B\right)\). The \textit{core} of a game is the set of additive set functions that exceed \(\upsilon \) pointiwse, i.e., \[ \operatorname{core}\left( \upsilon \right) :=\left\{ m\text{ additive s.t. }m\left(A\right) \geq \upsilon \left( A\right) ,\forall A\in \mathcal{A}\text{ and } m\left( \Omega \right) =\upsilon \left( \Omega \right) \right\}. \] It is assumed that \(\left( \Omega ,\mathcal{A}\right) \) is a measurable space and \textsl{P}\(:=\left( P_{1},\dots,P_{N}\right) \) is a vector of \textsl{N} non-atomic probability measures on \(\left( \Omega ,\mathcal{A} \right).\) A \textit{measure} game is a game of the form \(\upsilon = f\left( \text{P}\right) \), where \(f:\left[ 0,1\right] ^{N}\rightarrow\mathbb{R}\) satisfies \(f(0,\dots,0) =0\). It is assumed that \(f\) is continuous and it is \textit{ultramodular}, i.e., \(f\left( x+h\right) -f\left( x\right) \leq f\left( y+h\right) -f\left( y\right) \) for all \(\left( x,y\right) \in \left[ 0,1\right] ^{N}\) with \(x\leq y\) and \(h\geq 0\) such that \(\left( x+h,y+h\right) \in \left[ 0,1\right] ^{2N}\). It is known that for \(f\) continuous and ultramodular the game \(\upsilon \) is convex, that any element of \(\operatorname{core}\left( f\left( P\right) \right) \) is a measure and any measure \(m\in \operatorname{core}\left( f\left( P\right) \right) \) is absolutely continuous with respect to the probability \(\overline{P}\) defined by: \(\overline{P}:=\frac{1}{N}\sum_{i=1}^{N}P_{i}\). Given a probability space \(\left( \Omega _{1},\mathcal{A}_{1},Q_{1}\right) \), a measurable space \(\left( \Omega _{2},\mathcal{A}_{2}\right) \) and a measuarable map \(Y:\Omega _{1}\rightarrow \Omega _{2},\) the \textit{push-forward} of \(Q_{1}\) through \(Y,\) denoted by \(Y\#Q_{1}\), is the probability measure on \(\left( \Omega _{2},\mathcal{A}_{2}\right) \) defined by: \(Y\#Q_{1}\left( B\right) :=Q_{1}\left( Y^{-1}\left( B\right) \right) \) for every \(B\in \mathcal{A}_{2}\). Given \(X\in L^{\infty }\left(\Omega ,\mathcal{A},\overline{P}\right)\), define the distribution functions \(G_{X}^{1},\dots,G_{X}^{N}\) and \(\overline{G}_{X}\) by: \[ G_{X}^{i}\left( u\right) :=P_{i}\left( \left\{ X\geq u\right\} \right),i=1,\dots,N; \overline{G}_{X}\left( u\right) := \overline{P}\left( \left\{X\geq u\right\} \right) ,\forall u\in\mathbb{R}. \] Also define \(G_{X}:=\left( G_{X}^{1},\dots,G_{X}^{N}\right) \) and the \textit{nonincreasing rearrangement} of \(X\) with respect to \(P_{i}\) (or the \textit{generalized inverse} of \(G_{X}^{i}\)), \(\left( G_{X}^{i}\right) ^{-1}\) by: \(\left( G_{X}^{i}\right) ^{-1}\left( t\right) :=\sup \left\{ u\in\mathbb{R} :\left( G_{X}^{i}\right) \left( u\right) \geq t\right\} ,\forall t\in \left(0,1\right)\). For \(f\) continuous and ultramodular, define the set \(K_{f}:=\left\{\left( \varphi_{1},\dots,\varphi _{N}\right), \varphi _{i}:\left[0,1\right] \rightarrow\mathbb{R} \text{ s.t. }\sum_{i=1}^{N}\varphi _{i}\left( \gamma _{i}\right) \geq f\left( \gamma \right) ,\forall \gamma \in \left[ 0,1\right]^{N}\right\}\). Let \(X\in L^{\infty }\left( \Omega ,\mathcal{A},\overline{P}\right) \) be s.t. \(G_{X}:=\left( G_{X}^{1},\dots,G_{X}^{N}\right) \) is continuous on \(\mathbb{R}\). A collection of functions \(\left( \varphi_{1},\dots,\varphi_{N}\right)\), with each \(\varphi_{i}: \left[0,1\right] \rightarrow\mathbb{R},\) is called a \textit{Kantorovich potential } associated to \(G_{X}\) if and only if \(\left( \varphi_{1},\dots,\varphi_{N}\right) \in K_{f}\) and : \(\sum_{i=1}^{N}\varphi _{i}\left( G_{X}^{i}\left( u\right) \right) = f\left( G_{X}\left( u\right) \right) ,\forall u\in\mathbb{R}\). The following representation theorem for the core is proved. Theorem: Let \(f\in C^{0}\left( \left[ 0,1\right] ^{N}, \mathbb{R}\right)\) be ultramodular and satisfy \(f(0,\dots,0) = 0\) then: \[ \operatorname{core}\left( f\left( P\right) \right) =\overline{co}\left\{\sum_{i=1}^{N}D_{i}^{+}f\left( G_{X}\left( X\right) \right) \frac{dP_{i}}{d\overline{P}},X\in L^{\infty }\left( \Omega ,\mathcal{A},\overline{P}\right) ,s.t.G_{X}\text{ is continuous}\right\}, \] where \(\frac{dP_{i}}{d\overline{P}}\) denotes the Radon-Nikodym of \(P_{i}\) with respect to \(\overline{P}\) and \(\overline{co}\) denotes closed convex hull with respect to the norm topology of \(L^{1}\left( \Omega ,\mathcal{A},\overline{P}\right).\)